mirror of
https://ghfast.top/https://github.com/aeroxw/tick-stock-panel.git
synced 2026-09-12 15:34:16 +08:00
feat(platform): 因子平台与因子↔策略双向联动 v0.2.3
- 因子平台: /factors 一级页(检验/因子库/编辑器/组合/挖掘), DSL 公式因子(25 算子点选、双语字段、我的因子模板、脏公式守卫), 版本与生命周期, 自动挖掘 L1 统计筛选 - 因子↔策略四条桥: 触发器 Zap 快建因子条件信号、因子一键生成排名策略、自定义信号 AI 提示词接入因子分组、策略回测因子归因(胜/败单入场信号日因子均值, 独立 tab, 双语因子名) - 回测: 统计卡新增盈亏比(≥1 红/<1 绿), 蒙卡回撤合并为中位/95% 双值卡(自适应字号), 高级设置基础过滤与策略编辑器参数对齐(5 组区间) - 信号库独立页 /signals(原设置 tab 迁出), 持仓提醒入导航; 挖掘并入因子页第 5 tab, /mining 旧链接重定向 - 研究线配套: 因子目录 61→77(评分/矩阵双内核), stats_v2(Newey-West/BH-FDR/DSR), enriched 管道与异动/报价服务配套调整 - 文档: README 导航与特性表、features.md 因子平台章节、操作说明书 9.2、factor-platform-plan 执行状态与 §5、二开文档桥接说明; 交流与支持节改版 - 版本 0.2.2 → 0.2.3; 后端全量 1625 passed(1 例环境性跳过), 前端 build 通过
This commit is contained in:
shy3130
@@ -45,7 +45,8 @@
|
||||
| 🔀 **能力路由** | 多数据集(日K/除权/实时/分钟/盘口/财务,持续扩展)按源能力独立路由,任选组合 | [custom-data-source.md](./docs/custom-data-source.md) |
|
||||
| 🔍 **选股引擎** | 25 个内置策略 + 分钟策略 + 自定义信号 + AI 生成,Polars 毫秒级扫全 A 股 | [strategy.md](./docs/strategy.md) |
|
||||
| 📊 **指标流水线** | MA/EMA/MACD/RSI/KDJ/布林/量比等 68 列指标与信号,一次扫表落盘 enriched Parquet | [features.md](./docs/features.md) |
|
||||
| 🧪 **回测研究** | 因子/策略/分钟回测 + 财务快照因子(点时口径),T+1/费用/滑点约束,结果可导出 | [features.md](./docs/features.md) |
|
||||
| 🧪 **回测研究** | 因子/策略/分钟回测 + 财务快照因子(点时口径),T+1/费用/滑点约束,评分策略附带因子归因 | [features.md](./docs/features.md) |
|
||||
| 🔬 **因子平台** | DSL 自定义因子(编辑器 25 算子点选/试算/版本) + 检验/组合,与策略双向联动(一键生成策略/触发器引用因子/回测归因) | [factor-platform-plan.md](./docs/factor-platform-plan.md) |
|
||||
| ⛏️ **因子挖掘** | 嵌套样本外搜索多因子排名组合,与自有策略对照,候选库显式发布、永不自动上线 | [mining.md](./docs/mining.md) |
|
||||
| 🌡️ **市场环境** | 情绪周期 6 阶段(连板梯队驱动)+ 概念/行业主线排名,与 5 档环境分并存 | [market-phase.md](./docs/market-phase.md) |
|
||||
| 🚨 **异动监控** | 竞价/盘中/偏移三类异动一页覆盖:同花顺风向标 + 当日信号聚合 + 交易所偏离值口径 | — |
|
||||
@@ -66,10 +67,11 @@
|
||||
- **策略** Screener — Polars 毫秒级扫描全 A 股,日线/分钟策略统一单池,按策略声明周期自动路由执行
|
||||
- **回测** Backtest — 四种研究视图:
|
||||
- **因子回测** — IC/IR、分层收益、多空组合,62+ 因子目录先筛掉无效指标
|
||||
- **策略回测** — 净值曲线、回撤、夏普、胜率,T+1/手续费/滑点/止损,SSE 流式进度
|
||||
- **策略回测** — 净值曲线、回撤、夏普、胜率、盈亏比、蒙卡回撤,T+1/手续费/滑点/止损,SSE 流式进度;评分因子策略附带「因子归因」(胜/败单入场信号日因子对比)
|
||||
- **分钟策略回测** — 逐交易日回放信号、分钟收盘入场,分钟级成交明细
|
||||
- **验证** — 参数敏感性与滚动样本外
|
||||
- 研究闭环:结果导出 CSV(概要/净值/交易明细/分标的统计) → 保存候选 → **一键载入复测**
|
||||
- **因子** Factors — 检验/因子库/编辑器/组合四 tab:IC·分层·Newey-West 检验、自定义 DSL 因子(25 算子点选、双语字段、我的因子模板)、版本与生命周期管理;因子库可**一键生成排名策略**,策略触发器可直接引用因子条件
|
||||
- **挖掘** Mining — 嵌套样本外因子与策略挖掘:训练区间因子方向重估 + 相关性去重 + 多因子排名组合搜索,自有策略作对照轨;候选入库,显式确认后才发布,永不自动上线
|
||||
|
||||
**📈 个股与板块分析**
|
||||
@@ -81,6 +83,8 @@
|
||||
|
||||
**🔔 监控与复盘**
|
||||
- **监控中心** Monitor — 策略/个股信号/价格/异动四类规则,支持自选分组作用域,盘中实时弹窗 + 语音播报(播报个股名称与信号) + 触发记录持久化
|
||||
- **持仓提醒** Lots — 记录个股/ETF 买入批次,自动生成止盈止损/到期监控规则
|
||||
- **信号库** Signals — 内置预计算信号 + 自定义条件信号(含因子条件与 AI 生成),供策略触发器/回测/监控统一取用
|
||||
- **异动监控** Abnormal Moves — 按交易时间线三 tab:
|
||||
- **竞价异动** — 同花顺盘前风向标(含当日/次日真实收益对照、追高风险标记)+ 全市场竞价扫描(待采集任务)
|
||||
- **盘中异动** — 涨停/炸板/翘板/跌停/新高/新低/放量当日信号聚合,零新增采集
|
||||
@@ -90,7 +94,7 @@
|
||||
**🗄️ 数据与扩展**
|
||||
- **数据** Data — 本地数据画像与同步状态(维表/日K/除权/Enriched/指数/ETF/分钟K/财务),盘后管道与历史扩展
|
||||
- **扩展分析** (动态菜单) — 把任意第三方/扩展数据字段配成一级菜单,与内置数据同台分析
|
||||
- **设置** Settings — 数据源与能力检测(能力路由矩阵、档位徽章)、AI 接口、实时监控、扩展页面、信号库、菜单与系统设置
|
||||
- **设置** Settings — 数据源与能力检测(能力路由矩阵、档位徽章)、AI 接口、实时监控、扩展页面、菜单与系统设置
|
||||
|
||||
</details>
|
||||
|
||||
@@ -346,37 +350,32 @@ fork同时请点个star哦,欢迎 Issue 和 PR。
|
||||
|
||||
---
|
||||
|
||||
## 💬 交流群
|
||||
|
||||
欢迎加入交流群,一起讨论交流。作者个人维护的部分个性化接口,统一公布在群公告中,供大家免费使用。
|
||||
|
||||
<img src="./community-qr-code.jpg" alt="交流群二维码" width="240" />
|
||||
|
||||
---
|
||||
|
||||
## ❤️ 支持项目
|
||||
## ❤️ 支持项目 / 💬 交流群
|
||||
|
||||
<div align="center">
|
||||
|
||||
如果这个项目对你有帮助,欢迎请作者喝杯咖啡 ☕
|
||||
|
||||
<table>
|
||||
<tr>
|
||||
<td width="50%" align="center"><b>微信赞赏</b></td>
|
||||
<td width="50%" align="center"><b>支付宝</b></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td width="50%" align="center"><img src="./assets/support/wechat-appreciation.jpg" alt="微信赞赏码 · 感谢道友支持 愿一路长红" height="280" /></td>
|
||||
<td width="50%" align="center"><img src="./assets/support/alipay.jpg" alt="支付宝收款码 · 打开支付宝扫一扫" height="280" /></td>
|
||||
<td width="50%" align="center">
|
||||
<b>❤️ 支持项目</b><br/>
|
||||
<sub>如果这个项目对你有帮助,欢迎请作者喝杯咖啡 ☕</sub>
|
||||
<table>
|
||||
<tr><td align="center"><img src="./assets/support/wechat-appreciation.jpg" alt="微信赞赏码 · 感谢道友支持 愿一路长红" height="280" /></td></tr>
|
||||
<tr><td align="center"><sub>愿道友一路长红 📈</sub></td></tr>
|
||||
</table>
|
||||
</td>
|
||||
<td width="50%" align="center">
|
||||
<b>💬 交流群</b><br/>
|
||||
<sub>欢迎加入交流群,一起讨论交流<br/>个性化接口统一公布在群公告,免费使用</sub>
|
||||
<table>
|
||||
<tr><td align="center"><img src="./community-qr-code.jpg" alt="交流群二维码 · 个人维护的个性化接口见群公告" height="280" /></td></tr>
|
||||
</table>
|
||||
</td>
|
||||
</tr>
|
||||
</table>
|
||||
|
||||
愿道友一路长红 📈
|
||||
|
||||
</div>
|
||||
|
||||
> 打赏完全自愿,金额不限;不用于购买任何功能、数据权限、投资建议
|
||||
>
|
||||
> 作者精力有限,优先响应赞助回馈,希望理解
|
||||
|
||||
---
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
|
||||
import sys
|
||||
|
||||
__version__ = "0.2.2"
|
||||
__version__ = "0.2.3"
|
||||
|
||||
# Windows 默认 stdout/stderr 编码为 GBK(cp936),TickFlow SDK 内部输出含 emoji 的
|
||||
# 指数/标的名称(如 \U0001f193)时会抛 UnicodeEncodeError,导致请求失败。
|
||||
|
||||
@@ -128,9 +128,9 @@ class FactorColumnsResponse(BaseModel):
|
||||
|
||||
@router.get("/factor/columns")
|
||||
def factor_columns():
|
||||
"""返回可用的因子列列表。"""
|
||||
from app.backtest.factor import FACTOR_COLUMNS
|
||||
return {"columns": FACTOR_COLUMNS}
|
||||
"""返回可用的因子列列表 (含运行期注册的自定义/复合因子)。"""
|
||||
from app.factors.registry import factor_columns_view
|
||||
return {"columns": factor_columns_view()}
|
||||
|
||||
|
||||
class FactorBacktestRequest(BaseModel):
|
||||
@@ -149,9 +149,10 @@ class FactorBacktestRequest(BaseModel):
|
||||
@router.post("/factor/run")
|
||||
def factor_run(req: FactorBacktestRequest, request: Request):
|
||||
"""因子回测 — IC/IR 分析 + 分层回测。"""
|
||||
from app.backtest.factor import FACTOR_COLUMNS, FactorBacktestService, FactorConfig
|
||||
from app.backtest.factor import FactorBacktestService, FactorConfig
|
||||
from app.factors.registry import factor_columns_view
|
||||
|
||||
if req.factor_name not in {item["id"] for item in FACTOR_COLUMNS}:
|
||||
if req.factor_name not in {item["id"] for item in factor_columns_view()}:
|
||||
raise HTTPException(status_code=400, detail=f"不支持的因子: {req.factor_name}")
|
||||
|
||||
engine = _get_engine(request)
|
||||
@@ -184,7 +185,7 @@ def factor_run(req: FactorBacktestRequest, request: Request):
|
||||
|
||||
|
||||
class FactorBatchRequest(BaseModel):
|
||||
factor_names: list[str] = Field(..., min_length=1, max_length=64)
|
||||
factor_names: list[str] = Field(..., min_length=1, max_length=96) # 目录 77 + 自定义余量
|
||||
symbols: list[str] | None = None
|
||||
start: date | None = None
|
||||
end: date | None = None
|
||||
@@ -200,13 +201,13 @@ class FactorBatchRequest(BaseModel):
|
||||
def factor_batch(req: FactorBatchRequest, request: Request):
|
||||
"""批量筛选因子, 同一批次只加载并计算一次数据面板。"""
|
||||
from app.backtest.factor import (
|
||||
FACTOR_COLUMNS,
|
||||
FactorBacktestService,
|
||||
FactorBatchConfig,
|
||||
)
|
||||
from app.factors.registry import factor_columns_view
|
||||
|
||||
factor_names = list(dict.fromkeys(req.factor_names))
|
||||
allowed = {item["id"] for item in FACTOR_COLUMNS}
|
||||
allowed = {item["id"] for item in factor_columns_view()}
|
||||
invalid = [name for name in factor_names if name not in allowed]
|
||||
if invalid:
|
||||
raise HTTPException(status_code=400, detail=f"不支持的因子: {', '.join(invalid)}")
|
||||
|
||||
@@ -0,0 +1,454 @@
|
||||
"""因子注册表 API — 因子库 (P1) + 公式校验/试算 (P2) + 自定义/复合因子 CRUD (P3)。"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date, timedelta
|
||||
|
||||
import polars as pl
|
||||
from fastapi import APIRouter, HTTPException, Query, Request
|
||||
from pydantic import BaseModel, Field
|
||||
|
||||
from app.factors import store
|
||||
from app.factors.dsl import FACTOR_COLUMN, compile_formula
|
||||
from app.factors.registry import all_factors, unregister_factor
|
||||
|
||||
router = APIRouter(prefix="/api/factors", tags=["factors"])
|
||||
|
||||
|
||||
@router.get("")
|
||||
def list_factors(asset_type: str | None = Query(default=None, pattern="^(stock|etf)$")) -> dict:
|
||||
"""注册表因子列表; asset_type 过滤适用资产 (财务因子仅股票)。"""
|
||||
specs = all_factors(asset_type=asset_type)
|
||||
return {
|
||||
"factors": [
|
||||
{
|
||||
"id": spec.id,
|
||||
"label": spec.label,
|
||||
"group": spec.group,
|
||||
"kind": spec.kind,
|
||||
"version": spec.version,
|
||||
"formula": spec.formula_text,
|
||||
"direction": spec.direction,
|
||||
"unit": spec.unit,
|
||||
"warmup_bars": spec.warmup_bars,
|
||||
"pit": spec.pit,
|
||||
"asset_types": sorted(spec.asset_types),
|
||||
"stability": spec.stability,
|
||||
"scale_free": spec.scale_free,
|
||||
"dependencies": sorted(spec.dependencies),
|
||||
}
|
||||
for spec in specs
|
||||
]
|
||||
}
|
||||
|
||||
|
||||
class FormulaValidateRequest(BaseModel):
|
||||
formula: str = Field(..., min_length=1, max_length=2000)
|
||||
|
||||
|
||||
def _compiled_payload(compiled) -> dict:
|
||||
return {
|
||||
"ok": compiled.ok,
|
||||
"errors": [error.to_dict() for error in compiled.errors],
|
||||
"dependencies": sorted(compiled.dependencies),
|
||||
"referenced_factors": sorted(compiled.referenced_factors),
|
||||
"warmup_bars": compiled.warmup_bars,
|
||||
"cross_sectional": compiled.cross_sectional,
|
||||
}
|
||||
|
||||
|
||||
@router.post("/validate")
|
||||
def validate_formula(req: FormulaValidateRequest) -> dict:
|
||||
"""公式校验: 语法/语义/窗口纪律/依赖推导, 编译期 fail-closed。"""
|
||||
return _compiled_payload(compile_formula(req.formula))
|
||||
|
||||
|
||||
class FormulaTrialRequest(FormulaValidateRequest):
|
||||
asset_type: str = Field(default="stock", pattern="^(stock|etf)$")
|
||||
days: int = Field(default=40, ge=20, le=120)
|
||||
|
||||
|
||||
@router.post("/trial")
|
||||
def trial_formula(req: FormulaTrialRequest, request: Request) -> dict:
|
||||
"""公式试算: 最近 N 个交易日截面 Rank IC 快照 (复用回测面板与虚拟因子物化路径)。"""
|
||||
compiled = compile_formula(req.formula)
|
||||
if not compiled.ok:
|
||||
raise HTTPException(status_code=400, detail={"errors": [error.to_dict() for error in compiled.errors]})
|
||||
|
||||
from app.api.backtest import _get_engine
|
||||
|
||||
# 交易日 → 自然日换算 (A股年均 243 交易日 ≈ 1.48 自然日/交易日), 留 buffer
|
||||
calendar_days = int((compiled.warmup_bars + req.days) * 1.6) + 15
|
||||
start = date.today() - timedelta(days=calendar_days)
|
||||
# 面板基础物理列 (load_panel 只返回 parquet 物理列, 因子列由补算路径生成)
|
||||
base_columns = ["symbol", "date", "open", "high", "low", "close", "volume", "amount", "turnover_rate"]
|
||||
if "consecutive_limit_ups" in compiled.dependencies:
|
||||
base_columns.append("consecutive_limit_ups")
|
||||
engine = _get_engine(request)
|
||||
panel = engine.load_panel(None, start, date.today(), columns=base_columns, asset_type=req.asset_type)
|
||||
if panel.is_empty():
|
||||
raise HTTPException(status_code=400, detail="当前数据目录无可用历史数据, 无法试算")
|
||||
|
||||
# 复用检验引擎同一条补算路径 (compute_indicators + 虚拟因子物化), 禁止第二套计算逻辑
|
||||
from app.backtest.factor import FactorBacktestService
|
||||
|
||||
physical = set(panel.columns)
|
||||
to_compute = set(compiled.referenced_factors) | {
|
||||
dep for dep in compiled.dependencies if dep not in physical
|
||||
}
|
||||
if to_compute:
|
||||
panel = FactorBacktestService._compute_missing_factors(panel, to_compute)
|
||||
|
||||
if compiled.frame_transform is None:
|
||||
raise HTTPException(status_code=500, detail="编译产物缺少帧变换")
|
||||
prepared = compiled.frame_transform(panel)
|
||||
if prepared is None:
|
||||
raise HTTPException(
|
||||
status_code=400,
|
||||
detail={"errors": [{
|
||||
"code": "E013", "message": "依赖列不可用: 面板缺少公式所需列",
|
||||
"position": {"offset": 0, "line": 1},
|
||||
"detail": {"missing": sorted((compiled.dependencies | compiled.referenced_factors) - set(panel.columns))},
|
||||
}]},
|
||||
)
|
||||
|
||||
total_rows = panel.height
|
||||
frame = (
|
||||
prepared
|
||||
.with_columns(
|
||||
(pl.col("close").shift(-1).over("symbol") / pl.col("close") - 1.0).alias("_next_return")
|
||||
)
|
||||
.filter(pl.col(FACTOR_COLUMN).is_not_null())
|
||||
.unique(subset=["symbol", "date"], keep="last")
|
||||
.sort(["symbol", "date"])
|
||||
)
|
||||
non_null_rows = frame.height
|
||||
if non_null_rows == 0:
|
||||
return {
|
||||
"ok": True, "n_dates": 0, "null_ratio": 1.0,
|
||||
"ic_mean": None, "ic_std": None, "ir": None, "ic_win_rate": None,
|
||||
"ic_series": [], "message": "试算区间内公式输出全为空 (检查预热窗口与数据范围)",
|
||||
}
|
||||
|
||||
ic_frame = (
|
||||
frame.filter(pl.col("_next_return").is_not_null())
|
||||
.group_by("date")
|
||||
.agg(
|
||||
pl.corr(pl.col(FACTOR_COLUMN).rank(method="average"), pl.col("_next_return").rank(method="average")).alias("ic"),
|
||||
pl.len().alias("n_symbols"),
|
||||
)
|
||||
.filter(pl.col("ic").is_not_null())
|
||||
.sort("date")
|
||||
.tail(req.days)
|
||||
)
|
||||
ic_series = [
|
||||
{"date": str(row["date"]), "ic": round(row["ic"], 4), "n_symbols": row["n_symbols"]}
|
||||
for row in ic_frame.to_dicts()
|
||||
]
|
||||
if ic_frame.is_empty():
|
||||
return {
|
||||
"ok": True, "n_dates": 0, "null_ratio": round(1.0 - non_null_rows / max(total_rows, 1), 4),
|
||||
"ic_mean": None, "ic_std": None, "ir": None, "ic_win_rate": None,
|
||||
"ic_series": [], "message": "无有效 IC 截面 (需每期 ≥2 只标的)",
|
||||
}
|
||||
stats = ic_frame.select(
|
||||
pl.col("ic").mean().alias("mean"),
|
||||
pl.col("ic").std(ddof=0).alias("std"),
|
||||
(pl.col("ic") > 0).mean().alias("win"),
|
||||
).row(0, named=True)
|
||||
ic_std = stats["std"]
|
||||
# Newey-West t (lag=1): 与检验页同源口径, 样本过少时不给 (fail-closed)
|
||||
t_newey_west = None
|
||||
if ic_frame.height >= 5:
|
||||
from app.backtest.stats_v2 import newey_west_t
|
||||
|
||||
values = ic_frame["ic"].to_numpy()
|
||||
nw = newey_west_t(values, lag=1)
|
||||
if nw is not None:
|
||||
t_newey_west = round(float(nw[0]), 3)
|
||||
return {
|
||||
"ok": True,
|
||||
"n_dates": ic_frame.height,
|
||||
"null_ratio": round(1.0 - non_null_rows / max(total_rows, 1), 4),
|
||||
"ic_mean": round(stats["mean"], 4),
|
||||
"ic_std": None if ic_std is None else round(ic_std, 4),
|
||||
"ir": None if not ic_std or ic_std == 0 else round(stats["mean"] / ic_std, 3),
|
||||
"ic_win_rate": round(stats["win"], 4),
|
||||
"t_newey_west": t_newey_west,
|
||||
"ic_series": ic_series,
|
||||
}
|
||||
|
||||
|
||||
# ── 自定义/复合因子 CRUD (P3) ──────────────────────────────
|
||||
|
||||
|
||||
class CustomFactorCreateRequest(BaseModel):
|
||||
id: str | None = Field(default=None, max_length=48)
|
||||
label: str = Field(..., min_length=1, max_length=32)
|
||||
group: str = Field(default="自定义", max_length=16)
|
||||
formula: str = Field(..., min_length=1, max_length=2000)
|
||||
description: str = Field(default="", max_length=500)
|
||||
direction: str = Field(default="none", pattern="^(high|low|none)$")
|
||||
|
||||
|
||||
class CompositeFactorCreateRequest(BaseModel):
|
||||
id: str | None = Field(default=None, max_length=48)
|
||||
label: str = Field(..., min_length=1, max_length=32)
|
||||
group: str = Field(default="组合", max_length=16)
|
||||
members: dict[str, float] = Field(..., min_length=2, max_length=8)
|
||||
description: str = Field(default="", max_length=500)
|
||||
direction: str = Field(default="none", pattern="^(high|low|none)$")
|
||||
|
||||
|
||||
def _data_dir(request: Request):
|
||||
from pathlib import Path
|
||||
|
||||
data_dir = getattr(getattr(request.app.state, "repo", None), "store", None)
|
||||
root = getattr(data_dir, "data_dir", None) if data_dir is not None else None
|
||||
if root is None:
|
||||
raise HTTPException(status_code=500, detail="数据目录不可用")
|
||||
return Path(root)
|
||||
|
||||
|
||||
def _slugify_id(label: str, prefix: str) -> str:
|
||||
base = "".join(ch if ch.isascii() and (ch.isalnum() or ch == "_") else "_" for ch in label.lower())
|
||||
candidate = f"{prefix}_{base}".strip("_")[:44]
|
||||
import re
|
||||
|
||||
candidate = re.sub(r"_+", "_", candidate)
|
||||
return candidate or f"{prefix}_f"
|
||||
|
||||
|
||||
def _resolve_id(requested: str | None, label: str, prefix: str) -> str:
|
||||
return requested.strip() if requested and requested.strip() else _slugify_id(label, prefix)
|
||||
|
||||
|
||||
def _next_version(data_dir, factor_id: str) -> int:
|
||||
for definition in store.load_all(data_dir):
|
||||
if str(definition.get("id")) == factor_id:
|
||||
return int(definition.get("version", 1)) + 1
|
||||
return 1
|
||||
|
||||
|
||||
def _trial_nonempty(request: Request, formula: str, asset_type: str = "stock") -> None:
|
||||
"""保存前置校验: 公式在最近 40 个交易日有非空输出 (设计 §3.5, fail-closed)。"""
|
||||
compiled = compile_formula(formula)
|
||||
if not compiled.ok:
|
||||
raise HTTPException(status_code=400, detail={"errors": [e.to_dict() for e in compiled.errors]})
|
||||
from app.api.backtest import _get_engine
|
||||
from app.backtest.factor import FactorBacktestService
|
||||
|
||||
calendar_days = int((compiled.warmup_bars + 40) * 1.6) + 15
|
||||
base_columns = ["symbol", "date", "open", "high", "low", "close", "volume", "amount", "turnover_rate"]
|
||||
engine = _get_engine(request)
|
||||
panel = engine.load_panel(None, date.today() - timedelta(days=calendar_days), date.today(), columns=base_columns, asset_type=asset_type)
|
||||
if panel.is_empty():
|
||||
raise HTTPException(status_code=400, detail="当前无历史数据, 无法完成保存前试算 (fail-closed)")
|
||||
physical = set(panel.columns)
|
||||
to_compute = set(compiled.referenced_factors) | {d for d in compiled.dependencies if d not in physical}
|
||||
if to_compute:
|
||||
panel = FactorBacktestService._compute_missing_factors(panel, to_compute)
|
||||
prepared = compiled.frame_transform(panel) if compiled.frame_transform else None
|
||||
if prepared is None or prepared[FACTOR_COLUMN].is_not_null().sum() == 0:
|
||||
raise HTTPException(status_code=400, detail="公式在最近 40 个交易日输出全为空, 拒绝保存")
|
||||
|
||||
|
||||
@router.post("/custom")
|
||||
def create_custom_factor(req: CustomFactorCreateRequest, request: Request) -> dict:
|
||||
"""保存自定义公式因子: 编译通过 + 服务端试算非空 (fail-closed)。"""
|
||||
data_dir = _data_dir(request)
|
||||
factor_id = _resolve_id(req.id, req.label, "uf")
|
||||
definition = {
|
||||
"id": factor_id,
|
||||
"kind": "custom",
|
||||
"version": _next_version(data_dir, factor_id),
|
||||
"label": req.label,
|
||||
"group": req.group,
|
||||
"formula": req.formula,
|
||||
"description": req.description,
|
||||
"direction": req.direction,
|
||||
"status": "draft",
|
||||
"created_at": store._now(),
|
||||
"updated_at": store._now(),
|
||||
}
|
||||
try:
|
||||
store.to_spec(definition) # 先做 schema/id/编译校验
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
_trial_nonempty(request, req.formula)
|
||||
try:
|
||||
store.register_definition(definition)
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
store.save_one(data_dir, definition)
|
||||
return {"ok": True, "id": factor_id, "version": definition["version"]}
|
||||
|
||||
|
||||
@router.post("/composite")
|
||||
def create_composite_factor(req: CompositeFactorCreateRequest, request: Request) -> dict:
|
||||
"""保存复合因子: 成员校验 + 循环引用检查 (无需试算, 值由成员物化路径计算)。"""
|
||||
data_dir = _data_dir(request)
|
||||
factor_id = _resolve_id(req.id, req.label, "cf")
|
||||
definition = {
|
||||
"id": factor_id,
|
||||
"kind": "composite",
|
||||
"version": _next_version(data_dir, factor_id),
|
||||
"label": req.label,
|
||||
"group": req.group,
|
||||
"members": req.members,
|
||||
"description": req.description,
|
||||
"direction": req.direction,
|
||||
"status": "draft",
|
||||
"created_at": store._now(),
|
||||
"updated_at": store._now(),
|
||||
}
|
||||
try:
|
||||
store.register_definition(definition)
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
store.save_one(data_dir, definition)
|
||||
return {"ok": True, "id": factor_id, "version": definition["version"]}
|
||||
|
||||
|
||||
class CustomFactorUpdateRequest(BaseModel):
|
||||
label: str = Field(..., min_length=1, max_length=32)
|
||||
group: str = Field(default="自定义", max_length=16)
|
||||
formula: str = Field(..., min_length=1, max_length=2000)
|
||||
description: str = Field(default="", max_length=500)
|
||||
direction: str = Field(default="none", pattern="^(high|low|none)$")
|
||||
|
||||
|
||||
@router.post("/custom/{factor_id}/update")
|
||||
def update_custom_factor(factor_id: str, req: CustomFactorUpdateRequest, request: Request) -> dict:
|
||||
"""编辑已有自定义因子: 编译校验 + 试算非空 (与创建同一门禁) → 版本提升注册。
|
||||
|
||||
公式变化时状态回 draft (生命周期语义: 编辑后需重新检验激活); 仅改名称/分组保留状态。
|
||||
"""
|
||||
data_dir = _data_dir(request)
|
||||
target = None
|
||||
for definition in store.load_all(data_dir):
|
||||
if str(definition.get("id")) == factor_id:
|
||||
target = definition
|
||||
break
|
||||
if target is None:
|
||||
raise HTTPException(status_code=404, detail=f"自定义因子不存在: {factor_id}")
|
||||
if str(target.get("kind", "custom")) != "custom":
|
||||
raise HTTPException(status_code=400, detail=f"仅自定义因子支持公式编辑 (kind={target.get('kind')})")
|
||||
formula_changed = str(target.get("formula")) != req.formula
|
||||
if formula_changed:
|
||||
_trial_nonempty(request, req.formula)
|
||||
target.update({
|
||||
"label": req.label,
|
||||
"group": req.group,
|
||||
"formula": req.formula,
|
||||
"description": req.description,
|
||||
"direction": req.direction,
|
||||
"version": int(target.get("version", 1)) + 1, # 版本提升 → 注册表允许覆盖
|
||||
"status": "draft" if formula_changed else str(target.get("status", "draft")),
|
||||
"updated_at": store._now(),
|
||||
})
|
||||
try:
|
||||
store.register_definition(target)
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
store.save_one(data_dir, target)
|
||||
return {"ok": True, "id": factor_id, "version": target["version"], "status": target["status"]}
|
||||
|
||||
|
||||
def _find_references(data_dir, factor_id: str) -> list[str]:
|
||||
"""扫描策略与复合因子定义中的引用 (删除前 fail-closed 检查)。"""
|
||||
references: list[str] = []
|
||||
strategies_dir = data_dir / "strategies"
|
||||
if strategies_dir.is_dir():
|
||||
for file in strategies_dir.glob("*.json"):
|
||||
try:
|
||||
text = file.read_text(encoding="utf-8")
|
||||
if factor_id in text:
|
||||
references.append(f"strategies/{file.name}")
|
||||
except OSError:
|
||||
continue
|
||||
for definition in store.load_all(data_dir):
|
||||
if str(definition.get("id")) == factor_id:
|
||||
continue
|
||||
members = definition.get("members")
|
||||
if isinstance(members, dict) and factor_id in members:
|
||||
references.append(f"custom_factors/{definition.get('id')}.json")
|
||||
return references
|
||||
|
||||
|
||||
@router.delete("/custom/{factor_id}")
|
||||
def delete_custom_factor(factor_id: str, request: Request, force: bool = Query(default=False)) -> dict:
|
||||
"""删除自定义/复合因子; 有引用时列出引用方并拒绝 (需 force)。"""
|
||||
data_dir = _data_dir(request)
|
||||
from app.factors.registry import get_factor
|
||||
|
||||
if get_factor(factor_id) is None and not store.delete_one(data_dir, factor_id):
|
||||
raise HTTPException(status_code=404, detail=f"因子不存在: {factor_id}")
|
||||
references = _find_references(data_dir, factor_id)
|
||||
if references and not force:
|
||||
raise HTTPException(
|
||||
status_code=409,
|
||||
detail={"message": "该因子仍有引用, 拒绝删除 (可带 force=true 强制)", "references": references},
|
||||
)
|
||||
try:
|
||||
unregister_factor(factor_id)
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
store.delete_one(data_dir, factor_id)
|
||||
return {"ok": True, "id": factor_id, "removed_references": references}
|
||||
|
||||
|
||||
class FactorStatusRequest(BaseModel):
|
||||
status: str = Field(..., pattern="^(draft|active|watch|retired)$")
|
||||
|
||||
|
||||
@router.post("/custom/{factor_id}/status")
|
||||
def update_factor_status(factor_id: str, req: FactorStatusRequest, request: Request) -> dict:
|
||||
"""生命周期状态迁移 (P4): draft->active->watch->retired, 编辑后回 draft。"""
|
||||
data_dir = _data_dir(request)
|
||||
target = None
|
||||
for definition in store.load_all(data_dir):
|
||||
if str(definition.get("id")) == factor_id:
|
||||
target = definition
|
||||
break
|
||||
if target is None:
|
||||
raise HTTPException(status_code=404, detail=f"自定义因子不存在: {factor_id}")
|
||||
target["status"] = req.status
|
||||
target["updated_at"] = store._now()
|
||||
try:
|
||||
# 动态因子先注销再注册: 元数据变更 (status/group) 不提升版本,
|
||||
# 直接 register 会因"版本未提升"被拒 (启动加载后的真实路径)
|
||||
unregister_factor(factor_id)
|
||||
store.register_definition(target) # 状态与 stability 联动
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
store.save_one(data_dir, target)
|
||||
return {"ok": True, "id": factor_id, "status": req.status}
|
||||
|
||||
|
||||
class FactorGroupRequest(BaseModel):
|
||||
group: str = Field(..., min_length=1, max_length=24)
|
||||
|
||||
|
||||
@router.post("/custom/{factor_id}/group")
|
||||
def update_factor_group(factor_id: str, req: FactorGroupRequest, request: Request) -> dict:
|
||||
"""修改单个自定义/复合因子的分组 (内置因子分组与快照/预设绑定, 不可改)。"""
|
||||
data_dir = _data_dir(request)
|
||||
group = req.group.strip()
|
||||
if not group:
|
||||
raise HTTPException(status_code=400, detail="分组名不能为空")
|
||||
target = None
|
||||
for definition in store.load_all(data_dir):
|
||||
if str(definition.get("id")) == factor_id:
|
||||
target = definition
|
||||
break
|
||||
if target is None:
|
||||
raise HTTPException(status_code=404, detail=f"自定义因子不存在: {factor_id}")
|
||||
target["group"] = group
|
||||
target["updated_at"] = store._now()
|
||||
try:
|
||||
unregister_factor(factor_id)
|
||||
store.register_definition(target)
|
||||
except ValueError as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
store.save_one(data_dir, target)
|
||||
return {"ok": True, "id": factor_id, "group": group}
|
||||
+125
-3
@@ -13,7 +13,6 @@ from fastapi import APIRouter, Header, HTTPException, Query, Request
|
||||
from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
|
||||
from sse_starlette.sse import EventSourceResponse
|
||||
|
||||
from app.backtest.factor import FACTOR_COLUMNS
|
||||
from app.backtest.mining import (
|
||||
MAX_BEAM_WIDTH,
|
||||
MAX_COMBINATION_SIZE,
|
||||
@@ -21,6 +20,7 @@ from app.backtest.mining import (
|
||||
evaluate_candidate_gate,
|
||||
)
|
||||
from app.enriched_generation import EnrichedGenerationUnavailableError
|
||||
from app.factors.registry import factor_columns_view
|
||||
from app.services import preferences
|
||||
from app.services.mining_jobs import (
|
||||
RUN_STATUSES,
|
||||
@@ -31,6 +31,7 @@ from app.services.mining_jobs import (
|
||||
MiningRunValidationError,
|
||||
)
|
||||
from app.services.mining_preflight import (
|
||||
enriched_partition_dates,
|
||||
mining_availability,
|
||||
require_mining_availability,
|
||||
)
|
||||
@@ -40,7 +41,9 @@ from app.services.mining_schedule import (
|
||||
)
|
||||
|
||||
router = APIRouter(prefix="/api/backtest/mining", tags=["backtest"])
|
||||
_FACTOR_IDS = frozenset(str(item["id"]) for item in FACTOR_COLUMNS)
|
||||
# 校验时动态读取 (含运行期注册的自定义/复合因子)
|
||||
def _known_factor_ids() -> frozenset[str]:
|
||||
return frozenset(str(item["id"]) for item in factor_columns_view())
|
||||
_MAX_ARTIFACT_BYTES = 64 * 1024 * 1024
|
||||
_SSE_POLL_SECONDS = 0.5
|
||||
_SSE_HEARTBEAT_SECONDS = 15.0
|
||||
@@ -87,7 +90,7 @@ class MiningStartRequest(BaseModel):
|
||||
@field_validator("factor_names")
|
||||
@classmethod
|
||||
def _known_factors(cls, values: list[str]) -> list[str]:
|
||||
unknown = sorted(set(values) - _FACTOR_IDS)
|
||||
unknown = sorted(set(values) - _known_factor_ids())
|
||||
if unknown:
|
||||
raise ValueError(f"unknown mining factors: {unknown}")
|
||||
return values
|
||||
@@ -119,6 +122,38 @@ class MiningSchedulePatch(BaseModel):
|
||||
mining_budget_profile: Literal["balanced", "strict"] | None = None
|
||||
|
||||
|
||||
class MiningAutoStartRequest(BaseModel):
|
||||
"""自动挖掘: 因子池由 L1 统计筛选自动生成, 不接受手动指定。"""
|
||||
|
||||
model_config = ConfigDict(extra="forbid", strict=True)
|
||||
|
||||
asset_type: Literal["stock", "etf"] = "stock"
|
||||
start: date | None = None
|
||||
end: date | None = None
|
||||
budget_profile: Literal["exploratory", "balanced", "strict"] = "balanced"
|
||||
commission_pct: float = Field(0.0002, ge=0.0, le=0.05, allow_inf_nan=False)
|
||||
stamp_tax_pct: float = Field(0.0005, ge=0.0, le=0.05, allow_inf_nan=False)
|
||||
slippage_bps: float = Field(5.0, ge=0.0, le=1000.0, allow_inf_nan=False)
|
||||
correlation_threshold: float = Field(0.75, gt=0.0, le=1.0, allow_inf_nan=False)
|
||||
force: bool = False
|
||||
|
||||
@field_validator("start", "end", mode="before")
|
||||
@classmethod
|
||||
def _iso_dates(cls, value: Any) -> Any:
|
||||
if isinstance(value, str):
|
||||
try:
|
||||
return date.fromisoformat(value)
|
||||
except ValueError as exc:
|
||||
raise ValueError("dates must use ISO YYYY-MM-DD format") from exc
|
||||
return value
|
||||
|
||||
@model_validator(mode="after")
|
||||
def _date_range(self) -> MiningAutoStartRequest:
|
||||
if self.start is not None and self.end is not None and self.start > self.end:
|
||||
raise ValueError("start must not be after end")
|
||||
return self
|
||||
|
||||
|
||||
@router.get("/availability")
|
||||
def get_availability(
|
||||
request: Request,
|
||||
@@ -237,6 +272,93 @@ def cancel_run(run_id: str, request: Request) -> dict[str, Any]:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
|
||||
|
||||
@router.post("/auto")
|
||||
def start_auto_run(payload: MiningAutoStartRequest, request: Request) -> dict[str, Any]:
|
||||
"""自动挖掘: L1 统计筛选全量因子 → 达标池 → 复用挖掘任务管理启动嵌套样本外验证。
|
||||
|
||||
筛选结果随请求持久化 (request.auto_screening), 供结果页展示达标因子清单与
|
||||
失败原因分布; 无达标因子时返回 started=false 而不是报错。
|
||||
"""
|
||||
from app.services.auto_mining import screen_all_factors
|
||||
|
||||
manager = _manager(request)
|
||||
data_dir = request.app.state.repo.store.data_dir
|
||||
try:
|
||||
require_mining_availability(
|
||||
data_dir,
|
||||
asset_type=payload.asset_type,
|
||||
budget_profile=payload.budget_profile,
|
||||
start=payload.start,
|
||||
end=payload.end,
|
||||
)
|
||||
engine = getattr(request.app.state, "backtest_engine", None)
|
||||
if engine is None:
|
||||
from app.backtest.engine import BacktestEngine
|
||||
|
||||
engine = BacktestEngine(request.app.state.repo)
|
||||
request.app.state.backtest_engine = engine
|
||||
all_dates = enriched_partition_dates(data_dir, payload.asset_type)
|
||||
screen_end = payload.end or (all_dates[-1] if all_dates else date.today())
|
||||
screening = screen_all_factors(
|
||||
engine,
|
||||
asset_type=payload.asset_type,
|
||||
start=payload.start,
|
||||
end=screen_end,
|
||||
profile=payload.budget_profile,
|
||||
)
|
||||
except (MiningRunValidationError, ValueError) as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
except EnrichedGenerationUnavailableError as exc:
|
||||
raise HTTPException(
|
||||
status_code=400,
|
||||
detail="行情数据正在更新(enriched 发布中), 请等数据更新完成后再开始挖掘",
|
||||
) from exc
|
||||
|
||||
if not screening["pool"]:
|
||||
return {"started": False, "reason": "no_qualified_factors", "screening": screening}
|
||||
|
||||
worker_request = {
|
||||
"factor_names": screening["pool"],
|
||||
"strategy_ids": [],
|
||||
"symbols": None,
|
||||
"asset_type": payload.asset_type,
|
||||
"start": payload.start.isoformat() if payload.start else None,
|
||||
"end": payload.end.isoformat() if payload.end else None,
|
||||
"budget_profile": payload.budget_profile,
|
||||
"commission_pct": payload.commission_pct,
|
||||
"stamp_tax_pct": payload.stamp_tax_pct,
|
||||
"slippage_bps": payload.slippage_bps,
|
||||
"correlation_threshold": payload.correlation_threshold,
|
||||
"max_combination_factors": 4,
|
||||
"beam_width": 12,
|
||||
"max_finalists": MAX_FINALISTS,
|
||||
"auto": True,
|
||||
"auto_screening": screening,
|
||||
}
|
||||
try:
|
||||
fingerprint = build_data_fingerprint(
|
||||
request.app.state.repo,
|
||||
request.app.state,
|
||||
worker_request,
|
||||
)
|
||||
manifest = manager.start(
|
||||
worker_request,
|
||||
fingerprint,
|
||||
force=payload.force,
|
||||
source="auto",
|
||||
)
|
||||
except (MiningRunValidationError, ValueError) as exc:
|
||||
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
||||
except EnrichedGenerationUnavailableError as exc:
|
||||
raise HTTPException(
|
||||
status_code=400,
|
||||
detail="行情数据正在更新(enriched 发布中), 请等数据更新完成后再开始挖掘",
|
||||
) from exc
|
||||
except MiningRunStoreError as exc:
|
||||
raise HTTPException(status_code=500, detail="failed to persist mining run") from exc
|
||||
return {"started": True, "run": _project_run(manager.store, manifest), "screening": screening}
|
||||
|
||||
|
||||
@router.get("/runs/{run_id}/result")
|
||||
def get_result(run_id: str, request: Request) -> dict[str, Any]:
|
||||
store = _manager(request).store
|
||||
|
||||
@@ -87,6 +87,24 @@ def get_options():
|
||||
groups.append({"key": cat, "label": label,
|
||||
"fields": [{"key": f, "label": ENRICHED_COLUMNS.get(f, f)} for f in cat_fields]})
|
||||
|
||||
# 注册表因子 (虚拟/自定义/复合): 历史路径由 compute_signals 复用评分物化
|
||||
# 管线补算; 已是物化列的基础因子 (rsi_14 等) 上面已分组, 此处跳过。
|
||||
from app.factors.registry import all_factors
|
||||
|
||||
factor_groups: dict[str, list[dict[str, str]]] = {}
|
||||
for spec in all_factors():
|
||||
if spec.id in allowed:
|
||||
continue
|
||||
label = spec.label
|
||||
if spec.warmup_bars > 1:
|
||||
label = f"{label} · 预热{spec.warmup_bars}日"
|
||||
if list(spec.asset_types) == ["stock"]:
|
||||
label = f"{label} · 仅股票"
|
||||
factor_groups.setdefault(spec.group or "因子", []).append({"key": spec.id, "label": label})
|
||||
for group_label, group_fields in factor_groups.items():
|
||||
groups.append({"key": f"factor:{group_label}", "label": f"因子 · {group_label}", "fields": group_fields})
|
||||
fields.extend(group_fields)
|
||||
|
||||
return {
|
||||
"fields": fields,
|
||||
"groups": groups,
|
||||
|
||||
@@ -17,12 +17,14 @@ from typing import Any, Literal
|
||||
import numpy as np
|
||||
import polars as pl
|
||||
|
||||
from app.backtest import stats_v2
|
||||
from app.backtest.engine import BacktestEngine
|
||||
from app.backtest.fundamentals import (
|
||||
FUNDAMENTAL_FACTOR_NAMES,
|
||||
attach_fundamental_factors,
|
||||
load_fundamental_snapshot,
|
||||
)
|
||||
from app.factors.registry import factor_columns_view as _factor_columns_view
|
||||
from app.strategy.scoring import (
|
||||
VIRTUAL_SCORING_DEPENDENCIES as DERIVED_FACTOR_DEPENDENCIES,
|
||||
)
|
||||
@@ -33,80 +35,8 @@ from app.strategy.scoring import (
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# 可研究因子目录。保留历史 ID 兼容已有候选方案; 价格尺度相关指标优先提供归一化版本。
|
||||
FACTOR_COLUMNS: list[dict] = [
|
||||
{"id": "momentum_5d", "label": "5日动量", "group": "动量", "desc": "5个交易日累计收益率"},
|
||||
{"id": "momentum_10d", "label": "10日动量", "group": "动量", "desc": "10个交易日累计收益率"},
|
||||
{"id": "momentum_20d", "label": "20日动量", "group": "动量", "desc": "20个交易日累计收益率"},
|
||||
{"id": "momentum_30d", "label": "30日动量", "group": "动量", "desc": "30个交易日累计收益率"},
|
||||
{"id": "momentum_60d", "label": "60日动量", "group": "动量", "desc": "60个交易日累计收益率"},
|
||||
{"id": "change_pct", "label": "日涨跌幅", "group": "动量", "desc": "当日收盘相对前收盘的收益率"},
|
||||
|
||||
{"id": "ma5_bias", "label": "MA5乖离", "group": "均线偏离", "desc": "收盘价 / MA5 - 1"},
|
||||
{"id": "ma10_bias", "label": "MA10乖离", "group": "均线偏离", "desc": "收盘价 / MA10 - 1"},
|
||||
{"id": "ma20_bias", "label": "MA20乖离", "group": "均线偏离", "desc": "收盘价 / MA20 - 1"},
|
||||
{"id": "ma30_bias", "label": "MA30乖离", "group": "均线偏离", "desc": "收盘价 / MA30 - 1"},
|
||||
{"id": "ma60_bias", "label": "MA60乖离", "group": "均线偏离", "desc": "收盘价 / MA60 - 1"},
|
||||
{"id": "ema5_bias", "label": "EMA5乖离", "group": "均线偏离", "desc": "收盘价 / EMA5 - 1"},
|
||||
{"id": "ema10_bias", "label": "EMA10乖离", "group": "均线偏离", "desc": "收盘价 / EMA10 - 1"},
|
||||
{"id": "ema20_bias", "label": "EMA20乖离", "group": "均线偏离", "desc": "收盘价 / EMA20 - 1"},
|
||||
{"id": "ema30_bias", "label": "EMA30乖离", "group": "均线偏离", "desc": "收盘价 / EMA30 - 1"},
|
||||
{"id": "ema60_bias", "label": "EMA60乖离", "group": "均线偏离", "desc": "收盘价 / EMA60 - 1"},
|
||||
|
||||
{"id": "rsi_6", "label": "RSI(6)", "group": "超买超卖", "desc": "6日相对强弱指标"},
|
||||
{"id": "rsi_14", "label": "RSI(14)", "group": "超买超卖", "desc": "14日相对强弱指标"},
|
||||
{"id": "rsi_24", "label": "RSI(24)", "group": "超买超卖", "desc": "24日相对强弱指标"},
|
||||
|
||||
{"id": "macd_hist", "label": "MACD柱(原值)", "group": "趋势", "desc": "兼容历史研究; 跨股票比较建议优先使用MACD柱强度"},
|
||||
{"id": "macd_dif_pct", "label": "MACD DIF强度", "group": "趋势", "desc": "MACD DIF / 收盘价"},
|
||||
{"id": "macd_dea_pct", "label": "MACD DEA强度", "group": "趋势", "desc": "MACD DEA / 收盘价"},
|
||||
{"id": "macd_hist_pct", "label": "MACD柱强度", "group": "趋势", "desc": "MACD柱 / 收盘价, 消除股价尺度影响"},
|
||||
{"id": "kdj_k", "label": "KDJ-K", "group": "趋势", "desc": "KDJ指标K值"},
|
||||
{"id": "kdj_d", "label": "KDJ-D", "group": "趋势", "desc": "KDJ指标D值"},
|
||||
{"id": "kdj_j", "label": "KDJ-J", "group": "趋势", "desc": "KDJ指标J值"},
|
||||
{"id": "boll_position", "label": "布林位置", "group": "趋势", "desc": "收盘价在布林带下轨到上轨之间的位置"},
|
||||
|
||||
{"id": "annual_vol_20d", "label": "20日波动率", "group": "波动率", "desc": "20日收益率年化标准差"},
|
||||
{"id": "atr_14", "label": "ATR(14)原值", "group": "波动率", "desc": "兼容历史研究; 跨股票比较建议优先使用ATR相对波动"},
|
||||
{"id": "atr_pct", "label": "ATR相对波动", "group": "波动率", "desc": "ATR(14) / 收盘价"},
|
||||
{"id": "amplitude", "label": "日振幅", "group": "波动率", "desc": "当日高低价差 / 前收盘价"},
|
||||
{"id": "boll_width", "label": "布林带宽", "group": "波动率", "desc": "布林带上下轨宽度 / MA20"},
|
||||
|
||||
{"id": "vol_ratio_5d", "label": "5日量比", "group": "量价", "desc": "当日成交量 / 前5日平均成交量"},
|
||||
{"id": "vol_ratio_10d", "label": "10日量比", "group": "量价", "desc": "当日成交量 / 前10日平均成交量"},
|
||||
{"id": "vol_trend_5_10", "label": "成交量趋势", "group": "量价", "desc": "5日平均成交量 / 10日平均成交量 - 1"},
|
||||
{"id": "turnover_rate", "label": "换手率", "group": "量价", "desc": "使用历史时点流通股本计算的当日换手率"},
|
||||
{"id": "turnover_ratio_5d", "label": "换手率放大", "group": "量价", "desc": "当日换手率 / 前5日平均换手率 - 1"},
|
||||
{"id": "log_amount", "label": "成交额对数", "group": "量价", "desc": "ln(成交额 + 1), 降低极端规模影响"},
|
||||
{"id": "amount_ratio_5d", "label": "成交额放大", "group": "量价", "desc": "当日成交额 / 前5日平均成交额 - 1"},
|
||||
|
||||
{"id": "gap_return", "label": "开盘跳空", "group": "价格位置", "desc": "开盘价 / 前收盘价 - 1"},
|
||||
{"id": "intraday_return", "label": "日内收益", "group": "价格位置", "desc": "收盘价 / 开盘价 - 1"},
|
||||
{"id": "close_position", "label": "收盘位置", "group": "价格位置", "desc": "收盘价在当日最低价到最高价之间的位置"},
|
||||
{"id": "distance_to_high_60d", "label": "距60日高点", "group": "价格位置", "desc": "收盘价 / 60日最高收盘价 - 1"},
|
||||
{"id": "distance_from_low_60d", "label": "距60日低点", "group": "价格位置", "desc": "收盘价 / 60日最低收盘价 - 1"},
|
||||
{"id": "vwap_bias", "label": "VWAP乖离", "group": "价格位置", "desc": "收盘价 / 当日成交均价 - 1, 成交均价 = 成交额 / (成交量x100)"},
|
||||
|
||||
{"id": "max_ret_20d", "label": "20日最大单日涨幅", "group": "收益形态", "desc": "近20个交易日单日涨幅最大值(彩票效应, 高值代表博彩型特征强)"},
|
||||
{"id": "ret_skew_20d", "label": "20日收益偏度", "group": "收益形态", "desc": "近20个交易日日收益偏度, 高值代表右偏(偶发大涨)"},
|
||||
{"id": "up_days_20d", "label": "20日上涨天数", "group": "收益形态", "desc": "近20个交易日中上涨天数(0~20)"},
|
||||
|
||||
{"id": "amihud_20d", "label": "20日Amihud非流动性", "group": "流动性", "desc": "近20日平均 |日涨跌幅| / 成交额(亿元), 高值代表流动性差"},
|
||||
{"id": "turnover_z_60d", "label": "换手率60日z分", "group": "流动性", "desc": "(当日换手率 - 前60日均值) / 前60日标准差, 衡量换手异动"},
|
||||
|
||||
{"id": "vol_price_corr_20d", "label": "20日量价相关", "group": "量价", "desc": "近20个交易日日涨跌幅与成交量的相关系数, 高值代表量价同向"},
|
||||
{"id": "vol_trend_5_60", "label": "量能趋势(5/60)", "group": "量价", "desc": "5日平均成交量 / 60日平均成交量 - 1"},
|
||||
|
||||
{"id": "limit_up_count_20d", "label": "涨停基因(20日)", "group": "涨停基因", "desc": "近20个交易日涨停次数"},
|
||||
{"id": "limit_up_count_60d", "label": "涨停基因(60日)", "group": "涨停基因", "desc": "近60个交易日涨停次数"},
|
||||
|
||||
{"id": "pb_latest", "label": "市净率(最新公告)", "group": "财务", "desc": "收盘价 / 最新已公告每股净资产; 无财务数据或公告前为空"},
|
||||
{"id": "roe_latest", "label": "ROE(最新公告)", "group": "财务", "desc": "最新已公告净资产收益率(%); 无财务数据或公告前为空"},
|
||||
{"id": "gross_margin_latest", "label": "毛利率(最新公告)", "group": "财务", "desc": "最新已公告销售毛利率(%)"},
|
||||
{"id": "net_margin_latest", "label": "净利率(最新公告)", "group": "财务", "desc": "最新已公告销售净利率(%)"},
|
||||
{"id": "revenue_yoy_latest", "label": "营收增速(最新公告)", "group": "财务", "desc": "最新已公告营业收入同比(%)"},
|
||||
{"id": "net_income_yoy_latest", "label": "净利增速(最新公告)", "group": "财务", "desc": "最新已公告归母净利润同比(%)"},
|
||||
{"id": "debt_ratio_latest", "label": "资产负债率(最新公告)", "group": "财务", "desc": "最新已公告资产负债率(%)"},
|
||||
]
|
||||
# P1 起目录元数据单一权威来源为 app/factors/registry.py, 本常量为兼容别名 (顺序与键不变)。
|
||||
FACTOR_COLUMNS: list[dict] = _factor_columns_view()
|
||||
|
||||
FACTOR_WARMUP_DAYS = 120
|
||||
FACTOR_METHODOLOGY_VERSION = "factor_v2"
|
||||
@@ -208,6 +138,12 @@ class FactorBatchItem:
|
||||
yearly_ic: list[dict] = field(default_factory=list)
|
||||
ic_decay: list[dict] = field(default_factory=list)
|
||||
regime_stats: list[dict] = field(default_factory=list)
|
||||
# metrics_v2 (P3): NW HAC t 值 (滞后=1, 日频 1 日前瞻) 与 BH-FDR q 值; 样本不足为 None
|
||||
t_naive: float | None = None
|
||||
t_newey_west: float | None = None
|
||||
nw_lag: int | None = None
|
||||
p_value: float | None = None
|
||||
q_value: float | None = None
|
||||
|
||||
|
||||
@dataclass
|
||||
@@ -357,6 +293,10 @@ class FactorBacktestService:
|
||||
**evaluate_kwargs,
|
||||
)
|
||||
long_short = result.long_short_stats
|
||||
# metrics_v2: 由 IC 序列推导 NW HAC t 值与 p 值 (日频 1 日前瞻 → 滞后 1)
|
||||
ic_values = [row.get("ic") for row in result.ic_series]
|
||||
t_naive = stats_v2.naive_t(ic_values)
|
||||
nw = stats_v2.newey_west_t(ic_values, lag=1)
|
||||
items.append(FactorBatchItem(
|
||||
factor_name=factor_name,
|
||||
label=str(meta.get("label", factor_name)),
|
||||
@@ -377,6 +317,14 @@ class FactorBacktestService:
|
||||
n_dates=result.n_dates,
|
||||
elapsed_ms=result.elapsed_ms,
|
||||
error=result.error,
|
||||
t_naive=t_naive,
|
||||
t_newey_west=nw[0] if nw else None,
|
||||
nw_lag=1 if nw else None,
|
||||
p_value=(
|
||||
stats_v2.normal_two_sided_p(nw[0]) if nw
|
||||
else stats_v2.normal_two_sided_p(t_naive) if t_naive is not None
|
||||
else None
|
||||
),
|
||||
))
|
||||
except Exception as exc: # 单因子失败不能中止整个筛选批次
|
||||
logger.exception("factor batch item failed: %s", factor_name)
|
||||
@@ -390,6 +338,10 @@ class FactorBacktestService:
|
||||
|
||||
n_symbols = max((item.n_symbols for item in items), default=0)
|
||||
n_dates = max((item.n_dates for item in items), default=0)
|
||||
# metrics_v2: 批内 BH-FDR q 值 (m = 可检验因子数, 计算失败项不计)
|
||||
q_values = stats_v2.bh_fdr_qvalues([item.p_value for item in items])
|
||||
for item, q_value in zip(items, q_values, strict=True):
|
||||
item.q_value = q_value
|
||||
return FactorBatchResult(
|
||||
run_id=run_id,
|
||||
config=result_config,
|
||||
@@ -701,8 +653,31 @@ class FactorBacktestService:
|
||||
logger.warning("factors %s cannot be computed, missing columns: %s", factor_cols, missing)
|
||||
return panel
|
||||
|
||||
from app.factors.registry import get_factor
|
||||
from app.indicators.pipeline import compute_indicators
|
||||
|
||||
# custom/composite 因子走注册表→DSL/组合 的同一条物化路径 (P3, 与策略评分共用);
|
||||
# 其底层依赖 (如 change_pct/ma20) 先经内置补算路径物化, 再做 DSL/组合物化。
|
||||
registry_names = {
|
||||
name for name in factor_cols
|
||||
if (spec := get_factor(name)) is not None and spec.kind in ("custom", "composite")
|
||||
}
|
||||
if registry_names:
|
||||
base_deps: set[str] = set()
|
||||
for name in registry_names:
|
||||
spec = get_factor(name)
|
||||
if spec is not None:
|
||||
base_deps.update(spec.dependencies)
|
||||
base_deps -= set(panel.columns) | registry_names
|
||||
if base_deps:
|
||||
panel = FactorBacktestService._compute_missing_factors(
|
||||
panel, base_deps, assume_sorted=assume_sorted,
|
||||
)
|
||||
panel = materialize_scoring_columns(panel, registry_names)
|
||||
factor_cols = factor_cols - registry_names
|
||||
if not factor_cols:
|
||||
return panel
|
||||
|
||||
derived = factor_cols & set(DERIVED_FACTOR_DEPENDENCIES)
|
||||
indicator_columns = factor_cols - derived
|
||||
for factor_name in derived:
|
||||
|
||||
@@ -3768,6 +3768,12 @@ _MATRIX_COMPUTED_FEATURES = frozenset({
|
||||
"amihud_20d", "turnover_z_60d", "vol_price_corr_20d",
|
||||
"vwap_bias", "vol_trend_5_60",
|
||||
"limit_up_count_20d", "limit_up_count_60d",
|
||||
# --- 扩充批次 (2026-09-05): 与注册表/scoring 口径一致的 16 个新虚拟因子 ---
|
||||
"log_float_mv", "mom_accel_20_60", "rsi_14_delta_5d",
|
||||
"overnight_ret_20d", "intraday_ret_20d", "downside_vol_20d",
|
||||
"vol_regime_5_60", "amplitude_trend_20_60", "obv_trend_20d",
|
||||
"amount_mean_20d", "turnover_mean_20d", "turnover_std_20d",
|
||||
"position_240d", "distance_to_high_240d", "kdj_kd_diff",
|
||||
})
|
||||
|
||||
|
||||
@@ -4016,6 +4022,81 @@ def _compute_matrix_feature(market: MarketDataMatrix, name: str) -> np.ndarray:
|
||||
hits = np.where(np.isfinite(consecutive) & (consecutive > 0), np.float32(1.0), np.float32(0.0))
|
||||
hits = hits.astype(np.float32)
|
||||
return valid_rolling_sum(hits, close_valid, window)
|
||||
# --- 扩充批次 (2026-09-05): numpy 内核实现, 口径与 strategy/scoring.py 一致 ---
|
||||
if name == "log_float_mv":
|
||||
turnover = market.field("turnover_rate")
|
||||
valid = close_valid & np.isfinite(turnover) & (turnover > 0) & (market.volume > 0)
|
||||
out = np.full(market.shape, np.nan, dtype=np.float32)
|
||||
np.multiply(market.close, market.volume, out=out, where=valid)
|
||||
np.divide(out, turnover, out=out, where=valid)
|
||||
np.log(out, out=out, where=valid)
|
||||
return out
|
||||
if name == "mom_accel_20_60" or name == "kdj_kd_diff":
|
||||
left, right = (
|
||||
(matrix_feature(market, "momentum_20d"), matrix_feature(market, "momentum_60d"))
|
||||
if name == "mom_accel_20_60"
|
||||
else (matrix_feature(market, "kdj_k"), matrix_feature(market, "kdj_d"))
|
||||
)
|
||||
out = np.full(market.shape, np.nan, dtype=np.float32)
|
||||
np.subtract(left, right, out=out, where=np.isfinite(left) & np.isfinite(right))
|
||||
return out
|
||||
if name == "rsi_14_delta_5d":
|
||||
rsi = matrix_feature(market, "rsi_14")
|
||||
return rsi - valid_shift(rsi, 5, np.isfinite(rsi))
|
||||
if name == "overnight_ret_20d":
|
||||
overnight = _matrix_relative(market.open, matrix_feature(market, "prev_close"))
|
||||
return valid_rolling_sum(overnight, np.isfinite(overnight), 20)
|
||||
if name == "intraday_ret_20d":
|
||||
intraday = _matrix_relative(market.close, market.open)
|
||||
return valid_rolling_sum(intraday, np.isfinite(intraday), 20)
|
||||
if name == "downside_vol_20d":
|
||||
daily = matrix_feature(market, "change_pct")
|
||||
downside = np.where(
|
||||
np.isfinite(daily), np.minimum(daily, np.float32(0.0)), np.nan,
|
||||
).astype(np.float32)
|
||||
mean_sq = valid_rolling_mean(np.square(downside, dtype=np.float32), np.isfinite(downside), 20)
|
||||
out = np.full(market.shape, np.nan, dtype=np.float32)
|
||||
np.sqrt(mean_sq, out=out, where=np.isfinite(mean_sq))
|
||||
return out
|
||||
if name == "vol_regime_5_60":
|
||||
daily = matrix_feature(market, "change_pct")
|
||||
valid = np.isfinite(daily)
|
||||
return _matrix_ratio(
|
||||
valid_rolling_std(daily, valid, 5, ddof=1),
|
||||
valid_rolling_std(daily, valid, 60, ddof=1),
|
||||
)
|
||||
if name == "amplitude_trend_20_60":
|
||||
amplitude = matrix_feature(market, "amplitude")
|
||||
valid = np.isfinite(amplitude)
|
||||
return _matrix_relative(
|
||||
valid_rolling_mean(amplitude, valid, 20),
|
||||
valid_rolling_mean(amplitude, valid, 60),
|
||||
)
|
||||
if name == "obv_trend_20d":
|
||||
daily = matrix_feature(market, "change_pct")
|
||||
volume_valid = close_valid & np.isfinite(market.volume)
|
||||
signed = np.where(
|
||||
np.isfinite(daily), np.sign(daily) * market.volume, np.nan,
|
||||
).astype(np.float32)
|
||||
total = valid_rolling_sum(signed, volume_valid & np.isfinite(daily), 20)
|
||||
scale = valid_rolling_mean(market.volume, volume_valid, 20) * np.float32(20.0)
|
||||
return _matrix_ratio(total, scale)
|
||||
if name == "amount_mean_20d":
|
||||
amount = market.field("amount")
|
||||
return valid_rolling_mean(amount / np.float32(1e8), np.isfinite(amount), 20)
|
||||
if name == "turnover_mean_20d" or name == "turnover_std_20d":
|
||||
turnover = market.field("turnover_rate")
|
||||
valid = np.isfinite(turnover)
|
||||
mean = valid_rolling_mean(turnover, valid, 20)
|
||||
if name == "turnover_mean_20d":
|
||||
return mean
|
||||
return _matrix_ratio(valid_rolling_std(turnover, valid, 20, ddof=1), mean)
|
||||
if name == "position_240d":
|
||||
high = valid_rolling_max(market.close, close_valid, 240)
|
||||
low = valid_rolling_min(market.close, close_valid, 240)
|
||||
return _matrix_ratio(market.close - low, high - low)
|
||||
if name == "distance_to_high_240d":
|
||||
return _matrix_relative(market.close, valid_rolling_max(market.close, close_valid, 240))
|
||||
raise ValueError(f"unsupported matrix feature: {name}")
|
||||
|
||||
|
||||
|
||||
@@ -17,7 +17,6 @@ import numpy as np
|
||||
import polars as pl
|
||||
|
||||
from app.backtest.factor import (
|
||||
FACTOR_COLUMNS,
|
||||
FACTOR_METHODOLOGY_VERSION,
|
||||
FACTOR_WARMUP_DAYS,
|
||||
FactorBacktestService,
|
||||
@@ -57,6 +56,7 @@ from app.enriched_generation import (
|
||||
EnrichedGenerationUnavailableError,
|
||||
enriched_publication_incomplete,
|
||||
)
|
||||
from app.factors.registry import factor_columns_view
|
||||
from app.services.mining_jobs import MiningRunStore
|
||||
from app.services.mining_preflight import enriched_partition_dates
|
||||
from app.services.mining_schedule import MINING_ALGORITHM_VERSION
|
||||
@@ -66,7 +66,6 @@ from app.strategy.engine import StrategyEngine
|
||||
ProgressCallback = Callable[[dict[str, Any]], None]
|
||||
CancelCheck = Callable[[], bool] | Any
|
||||
_PROFILE_NAMES = frozenset({"exploratory", "balanced", "strict"})
|
||||
_FACTOR_IDS = frozenset(str(item["id"]) for item in FACTOR_COLUMNS)
|
||||
_MINING_MATRIX_CACHE_BYTES = 32 * 1024 * 1024
|
||||
_RESULT_POLICY = BacktestResultPolicy(
|
||||
required_stats=frozenset({"total_return", "sharpe", "max_drawdown", "n_trades"}),
|
||||
@@ -807,7 +806,9 @@ def _decode_runtime_request(
|
||||
factor_names = tuple(str(value) for value in request.get("factor_names") or ())
|
||||
if not factor_names or len(set(factor_names)) != len(factor_names):
|
||||
raise ValueError("factor_names must be non-empty and unique")
|
||||
unknown_factors = sorted(set(factor_names) - _FACTOR_IDS)
|
||||
# 注册表动态读取: worker 子进程已在入口加载自定义/复合因子
|
||||
known_ids = frozenset(str(item["id"]) for item in factor_columns_view())
|
||||
unknown_factors = sorted(set(factor_names) - known_ids)
|
||||
if unknown_factors:
|
||||
raise ValueError(f"unknown mining factors: {unknown_factors}")
|
||||
if len(factor_names) > 48:
|
||||
@@ -1037,7 +1038,7 @@ def _build_artifacts(
|
||||
candidate.factor_names, candidate.directions, strict=True
|
||||
):
|
||||
direction_by_factor.setdefault(factor_name, int(direction))
|
||||
metadata = {str(item["id"]): item for item in FACTOR_COLUMNS}
|
||||
metadata = {str(item["id"]): item for item in factor_columns_view()}
|
||||
factor_rows = []
|
||||
for factor_name in request.factor_names:
|
||||
metric = latest_metrics[factor_name]
|
||||
|
||||
@@ -0,0 +1,147 @@
|
||||
"""metrics_v2 统计函数 (P3) — Newey-West HAC t 值 / BH-FDR q 值 / DSR。
|
||||
|
||||
运行时零新增第三方依赖 (后端无 scipy/statsmodels), 全部 numpy 手写;
|
||||
数值测试用固定黄金参考向量锁定 (tests/test_stats_v2.py)。
|
||||
|
||||
口径 (设计文档 factor-system-design.md §6):
|
||||
- IC 序列因 h 日前瞻收益存在 h-1 阶移动平均自相关, 主口径 t 值取 NW HAC, 滞后 L=h。
|
||||
- 多因子批量检验按 Benjamini-Hochberg 步进法控制 FDR。
|
||||
- DSR (Deflated Sharpe Ratio, Bailey & Lopez de Prado 2014) 用于多重试验校正后的
|
||||
夏普显著性; 期望最大夏普 EM = sqrt(V[SR]) * ((1-gamma)Φ^-1(1-1/N) + gammaΦ^-1(1-1/(Ne)))
|
||||
其中 gamma 为欧拉-马歇罗尼常数。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import math
|
||||
|
||||
import numpy as np
|
||||
|
||||
EULER_GAMMA = 0.5772156649015329
|
||||
|
||||
|
||||
def _clean_values(values) -> np.ndarray:
|
||||
array = np.asarray([value for value in values if value is not None and np.isfinite(value)], dtype=float)
|
||||
return array
|
||||
|
||||
|
||||
def newey_west_t(values, lag: int) -> tuple[float, float, float] | None:
|
||||
"""Newey-West HAC 稳健 t 统计量 (Bartlett 核)。
|
||||
|
||||
返回 (t值, 均值, NW标准误); 样本不足 (n <= lag+2) 或方差为零返回 None。
|
||||
"""
|
||||
array = _clean_values(values)
|
||||
n = array.size
|
||||
if n <= lag + 2 or n < 3:
|
||||
return None
|
||||
mean = float(array.mean())
|
||||
centered = array - mean
|
||||
# 长方差 S = gamma0 + 2 Σ_l w_l gamma_l, w_l = 1 - l/(lag+1) (Bartlett)
|
||||
gamma = [float(np.dot(centered[: n - lag_i], centered[lag_i:]) / n) for lag_i in range(lag + 1)]
|
||||
long_variance = gamma[0]
|
||||
for lag_i in range(1, lag + 1):
|
||||
weight = 1.0 - lag_i / (lag + 1)
|
||||
long_variance += 2.0 * weight * gamma[lag_i]
|
||||
long_variance = max(long_variance, 0.0)
|
||||
nw_se = math.sqrt(long_variance / n)
|
||||
if nw_se == 0:
|
||||
return None
|
||||
return (mean - 0.0) / nw_se, mean, nw_se
|
||||
|
||||
|
||||
def naive_t(values) -> float | None:
|
||||
array = _clean_values(values)
|
||||
n = array.size
|
||||
if n < 3:
|
||||
return None
|
||||
std = float(array.std(ddof=1))
|
||||
if std == 0:
|
||||
return None
|
||||
return float(array.mean()) / (std / math.sqrt(n))
|
||||
|
||||
|
||||
def normal_two_sided_p(t_stat: float) -> float:
|
||||
"""标准正态双侧 p 值: erfc(|t|/sqrt(2))。"""
|
||||
return math.erfc(abs(t_stat) / math.sqrt(2.0))
|
||||
|
||||
|
||||
def bh_fdr_qvalues(pvalues: list[float | None]) -> list[float | None]:
|
||||
"""Benjamini-Hochberg 步进法 q 值 (与输入等长, None 透传)。
|
||||
|
||||
m 取可检验假设数 (None 不计入); q_i = min over j>=rank_i { p_j * m / rank_j },
|
||||
从大到小单调回填保证递增约束。
|
||||
"""
|
||||
indexed = [
|
||||
(index, p) for index, p in enumerate(pvalues)
|
||||
if p is not None and np.isfinite(p)
|
||||
]
|
||||
qvalues: list[float | None] = [None] * len(pvalues)
|
||||
if not indexed:
|
||||
return qvalues
|
||||
m = len(indexed)
|
||||
indexed.sort(key=lambda pair: pair[1])
|
||||
running_min = float("inf")
|
||||
for reverse_rank in range(len(indexed) - 1, -1, -1):
|
||||
index, p = indexed[reverse_rank]
|
||||
rank = reverse_rank + 1
|
||||
candidate = p * m / rank
|
||||
running_min = min(running_min, candidate)
|
||||
qvalues[index] = min(1.0, running_min)
|
||||
return qvalues
|
||||
|
||||
|
||||
def _normal_ppf(probability: float) -> float:
|
||||
"""标准正态分位数 Acklam 逆逼近 (相对误差 < 1.15e-9), 零依赖替代 scipy.stats.norm.ppf。"""
|
||||
if not (0.0 < probability < 1.0):
|
||||
raise ValueError("probability 必须在 (0,1) 开区间")
|
||||
a = (-3.969683028665376e+01, 2.209460984245205e+02, -2.759285104469687e+02,
|
||||
1.383577518672690e+02, -3.066479806614716e+01, 2.506628277459239e+00)
|
||||
b = (-5.447609879822406e+01, 1.615858368580409e+02, -1.556989798598866e+02,
|
||||
6.680131188771972e+01, -1.328068155288572e+01)
|
||||
c = (-7.784894002430293e-03, -3.223964580411365e-01, -2.400758277161838e+00,
|
||||
-2.549732539343734e+00, 4.374664141464968e+00, 2.938163982698783e+00)
|
||||
d = (7.784695709041462e-03, 3.224671290700398e-01, 2.445134137142996e+00,
|
||||
3.754408661907416e+00)
|
||||
p_low, p_high = 0.02425, 1 - 0.02425
|
||||
if probability < p_low:
|
||||
q_value = math.sqrt(-2 * math.log(probability))
|
||||
return (((((c[0] * q_value + c[1]) * q_value + c[2]) * q_value + c[3]) * q_value + c[4]) * q_value + c[5]) / \
|
||||
((((d[0] * q_value + d[1]) * q_value + d[2]) * q_value + d[3]) * q_value + 1)
|
||||
if probability <= p_high:
|
||||
q_value = probability - 0.5
|
||||
r = q_value * q_value
|
||||
return (((((a[0] * r + a[1]) * r + a[2]) * r + a[3]) * r + a[4]) * r + a[5]) * q_value / \
|
||||
(((((b[0] * r + b[1]) * r + b[2]) * r + b[3]) * r + b[4]) * r + 1)
|
||||
q_value = math.sqrt(-2 * math.log(1 - probability))
|
||||
return -(((((c[0] * q_value + c[1]) * q_value + c[2]) * q_value + c[3]) * q_value + c[4]) * q_value + c[5]) / \
|
||||
((((d[0] * q_value + d[1]) * q_value + d[2]) * q_value + d[3]) * q_value + 1)
|
||||
|
||||
|
||||
def expected_max_sharpe(n_trials: int, variance_sharpes: float) -> float:
|
||||
"""N 次独立试验的期望最大夏普 EM (方差>0 时); 单次试验不校正。"""
|
||||
if n_trials <= 1 or variance_sharpes <= 0:
|
||||
return 0.0
|
||||
z1 = _normal_ppf(1.0 - 1.0 / n_trials)
|
||||
z2 = _normal_ppf(1.0 - 1.0 / (n_trials * math.e))
|
||||
return math.sqrt(variance_sharpes) * ((1.0 - EULER_GAMMA) * z1 + EULER_GAMMA * z2)
|
||||
|
||||
|
||||
def deflated_sharpe_psr(
|
||||
sharpe: float,
|
||||
n_obs: int,
|
||||
skewness: float | None = None,
|
||||
kurtosis: float | None = None,
|
||||
expected_max_sharpe: float = 0.0,
|
||||
) -> float | None:
|
||||
"""Deflated Sharpe (PSR 对 EM 校正) 概率; 参数不足或退化返回 None。
|
||||
|
||||
PSR = Φ( (SR - SR*) * sqrt(n-1) / sqrt(1 - gamma3 SR + (gamma4-1)/4 SR^2) )
|
||||
"""
|
||||
if n_obs < 5 or not np.isfinite(sharpe):
|
||||
return None
|
||||
skewness = 0.0 if skewness is None else skewness
|
||||
kurtosis = 3.0 if kurtosis is None else kurtosis
|
||||
denominator = 1.0 - skewness * sharpe + (kurtosis - 1.0) / 4.0 * sharpe * sharpe
|
||||
if denominator <= 0:
|
||||
return None
|
||||
statistic = (sharpe - expected_max_sharpe) * math.sqrt(n_obs - 1) / math.sqrt(denominator)
|
||||
return 0.5 * (1.0 + math.erf(statistic / math.sqrt(2.0)))
|
||||
@@ -591,6 +591,7 @@ class StrategyBacktestResult:
|
||||
trades: list[dict] = field(default_factory=list)
|
||||
per_symbol_stats: list[dict] = field(default_factory=list)
|
||||
strategy_info: dict = field(default_factory=dict)
|
||||
factor_attribution: dict | None = None
|
||||
elapsed_ms: float = 0.0
|
||||
error: str | None = None
|
||||
|
||||
@@ -655,6 +656,56 @@ class BacktestResultPolicy:
|
||||
return {key: value for key, value in stats.items() if key in keep}
|
||||
|
||||
|
||||
def _factor_attribution_summary(
|
||||
snapshot: pl.DataFrame,
|
||||
trades: list,
|
||||
) -> dict | None:
|
||||
"""v1 因子归因: 入场信号日因子快照 x 成交盈亏, 对比盈利/亏损单因子均值。
|
||||
|
||||
snapshot 来自 _apply_score 物化的候选行 (与评分同一条计算管线), 模拟结束后
|
||||
按 (symbol, 信号日) 关联成交。快照缺失、无可关联行或因子列全空时返回 None,
|
||||
归因失败不影响回测主结果。
|
||||
"""
|
||||
factor_cols = [c for c in snapshot.columns if c not in ("symbol", "date")]
|
||||
if not factor_cols or not trades:
|
||||
return None
|
||||
normalized = snapshot.with_columns(
|
||||
pl.col("date").cast(pl.Utf8).str.slice(0, 10).alias("date")
|
||||
)
|
||||
symbols: list[str] = []
|
||||
days: list[str] = []
|
||||
pnls: list[float] = []
|
||||
for trade in trades:
|
||||
day = trade.entry_signal_date or trade.entry_date
|
||||
if day is None:
|
||||
continue
|
||||
symbols.append(trade.symbol)
|
||||
days.append(str(day)[:10])
|
||||
pnls.append(float(trade.pnl_pct))
|
||||
if not symbols:
|
||||
return None
|
||||
frame = pl.DataFrame({"symbol": symbols, "date": days, "pnl_pct": pnls})
|
||||
joined = frame.join(normalized, on=["symbol", "date"], how="left")
|
||||
win = joined.filter(pl.col("pnl_pct") > 0)
|
||||
lose = joined.filter(pl.col("pnl_pct") <= 0)
|
||||
factors: list[dict] = []
|
||||
for col in factor_cols:
|
||||
win_vals = win.get_column(col).drop_nulls().cast(pl.Float64)
|
||||
lose_vals = lose.get_column(col).drop_nulls().cast(pl.Float64)
|
||||
if win_vals.is_empty() and lose_vals.is_empty():
|
||||
continue
|
||||
factors.append({
|
||||
"factor": col,
|
||||
"win_mean": round(float(win_vals.mean()), 6) if not win_vals.is_empty() else None,
|
||||
"lose_mean": round(float(lose_vals.mean()), 6) if not lose_vals.is_empty() else None,
|
||||
"win_n": int(win_vals.len()),
|
||||
"lose_n": int(lose_vals.len()),
|
||||
})
|
||||
if not factors:
|
||||
return None
|
||||
return {"factors": factors, "n_win": win.height, "n_lose": lose.height}
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class PreparedMatrixBacktest:
|
||||
"""Job-scoped immutable market data reused by every optimizer trial."""
|
||||
@@ -1012,6 +1063,8 @@ class StrategyBacktestService:
|
||||
t0 = time.perf_counter()
|
||||
run_id = uuid.uuid4().hex[:10]
|
||||
result_policy = result_policy or BacktestResultPolicy()
|
||||
# 因子归因快照容器: 日线路径在 _apply_score 里填充, 其余路径保持空
|
||||
factor_snapshot: dict = {}
|
||||
|
||||
def _err(msg: str) -> StrategyBacktestResult:
|
||||
return StrategyBacktestResult(
|
||||
@@ -1505,7 +1558,7 @@ class StrategyBacktestService:
|
||||
|
||||
candidate_filter_mask = self._build_candidate_filter_mask(panel, s, params)
|
||||
candidate_mask = basic_mask & candidate_filter_mask
|
||||
panel = self._apply_score(panel, s, overrides, universe_mask=candidate_mask)
|
||||
panel = self._apply_score(panel, s, overrides, universe_mask=candidate_mask, factor_snapshot=factor_snapshot)
|
||||
formal_candidate_mask = candidate_mask & formal_range
|
||||
entry_mask = self._build_entry_mask_from_candidate(panel, candidate_mask, s, entry_signals)
|
||||
entry_mask = entry_mask & formal_range
|
||||
@@ -1662,6 +1715,16 @@ class StrategyBacktestService:
|
||||
|
||||
selected_stats = result_policy.select_stats(result.stats)
|
||||
|
||||
# 因子归因 (fail-open): 快照与成交按信号日关联, 失败只记日志不影响结果
|
||||
factor_attribution = None
|
||||
if factor_snapshot and result.trades and result_policy.include_trades:
|
||||
try:
|
||||
factor_attribution = _factor_attribution_summary(
|
||||
factor_snapshot["frame"], result.trades
|
||||
)
|
||||
except Exception as exc:
|
||||
logger.warning("factor attribution failed: %s", exc)
|
||||
|
||||
elapsed = (time.perf_counter() - t0) * 1000
|
||||
|
||||
return StrategyBacktestResult(
|
||||
@@ -1682,6 +1745,7 @@ class StrategyBacktestService:
|
||||
else []
|
||||
),
|
||||
strategy_info=strategy_info,
|
||||
factor_attribution=factor_attribution,
|
||||
elapsed_ms=round(elapsed, 1),
|
||||
)
|
||||
|
||||
@@ -2465,6 +2529,7 @@ class StrategyBacktestService:
|
||||
s: StrategyDef,
|
||||
overrides: dict | None,
|
||||
universe_mask: pl.Series | None = None,
|
||||
factor_snapshot: dict | None = None,
|
||||
) -> pl.DataFrame:
|
||||
scoring = effective_scoring(s.meta.get("scoring"), overrides)
|
||||
directions = effective_scoring_directions(overrides)
|
||||
@@ -2475,6 +2540,18 @@ class StrategyBacktestService:
|
||||
if has_universe:
|
||||
work = work.with_columns(universe_mask.rename("_score_universe"))
|
||||
|
||||
# 因子归因快照: 在临时因子列被 _finish 丢弃前, 截取候选行的
|
||||
# (symbol, date, 因子值)。与评分共用同一份物化结果, 无第二次计算。
|
||||
if factor_snapshot is not None:
|
||||
snapshot_cols = ["symbol", "date"] + [
|
||||
name for name in scoring if name in work.columns
|
||||
]
|
||||
if len(snapshot_cols) > 2:
|
||||
frame = work
|
||||
if has_universe:
|
||||
frame = frame.filter(pl.col("_score_universe"))
|
||||
factor_snapshot["frame"] = frame.select(snapshot_cols)
|
||||
|
||||
def _value_in_universe(value: pl.Expr) -> pl.Expr:
|
||||
if has_universe:
|
||||
return pl.when(pl.col("_score_universe")).then(value).otherwise(None)
|
||||
|
||||
@@ -182,6 +182,12 @@ def _worker_entry(task: dict[str, Any], event_queue, cancel_event) -> None:
|
||||
data_dir = Path(task["data_dir"])
|
||||
store = DataStore(data_dir)
|
||||
repo = KlineRepository(store)
|
||||
# 子进程不继承主进程的因子注册表; 自定义/复合因子 (uf_/cf_) 在任何
|
||||
# 涉及因子物化的 worker 任务里都依赖注册表, 启动时从存储加载。
|
||||
# 单个加载失败只跳过 (fail-open 跳过该因子), 与主进程启动行为一致。
|
||||
from app.factors.store import load_into_registry
|
||||
|
||||
load_into_registry(data_dir)
|
||||
strategy_engine = StrategyEngine(
|
||||
strategy_dirs=_strategy_dirs(data_dir),
|
||||
override_loader=lambda sid: strategy_config.load_override(data_dir, sid),
|
||||
|
||||
@@ -0,0 +1,749 @@
|
||||
"""因子公式 DSL 编译器 (P2)。
|
||||
|
||||
流水线: text → tokenizer → 递归下降解析(EBNF 见设计文档 §3.4) → AST → 语义检查
|
||||
→ 依赖/预热推导 → Polars Expr。编译失败返回结构化错误 (E001-E016), 不抛裸异常。
|
||||
|
||||
窗口纪律 (Polars 嵌套窗口会静默产出全 null, 必须在编译期杜绝):
|
||||
- 所有 ts_* 算子只向后看 (负 shift 常量层强制 E005)。
|
||||
- 时序子树仅在离开时序上下文时挂一次 over("symbol"); 截面算子挂 over("date")。
|
||||
- 截面算子消费含窗口的子树时, 编译为两阶段: 先把该子树物化为临时列 (单层 over),
|
||||
再对临时列做截面运算 —— frame_transform 负责按依赖顺序执行全部阶段。
|
||||
- 截面算子嵌在时序窗口内 (如 ts_mean(rank(x), n)) v1 不支持, 编译期 E009 拒绝。
|
||||
- 引用的注册因子(含 virtual)不内联表达式: 调用方用 materialize_scoring_columns
|
||||
物化成列, 编译产物统一以 pl.col(name) 引用; 运行期缺列即 fail-closed。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import copy
|
||||
import re
|
||||
from dataclasses import dataclass, field
|
||||
from functools import lru_cache
|
||||
from typing import Any
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.factors.registry import factor_dependencies, get_factor
|
||||
|
||||
FACTOR_COLUMN = "__dsl_factor__"
|
||||
|
||||
# 基准列 (设计文档 §3.1); 指标列 = 注册表 base 因子, 已注册因子 id 经注册表解析。
|
||||
BASE_COLUMNS: frozenset[str] = frozenset({
|
||||
"open", "high", "low", "close", "volume", "amount",
|
||||
"turnover_rate", "prev_close", "raw_close",
|
||||
})
|
||||
|
||||
MAX_AST_DEPTH = 12
|
||||
MAX_TOKENS = 200
|
||||
WINDOW_MIN, WINDOW_MAX = 2, 512
|
||||
DELAY_MAX = 512
|
||||
POWER_ABS_MAX = 4.0
|
||||
WINSORIZE_K_RANGE = (1.0, 6.0)
|
||||
|
||||
# 算子表: 名 -> (表达式参数个数, 常量参数名元组); 常量参数必须是数字字面量 (E003)。
|
||||
OPERATORS: dict[str, tuple[int, tuple[str, ...]]] = {
|
||||
"ts_mean": (1, ("n",)),
|
||||
"ts_std": (1, ("n",)),
|
||||
"ts_sum": (1, ("n",)),
|
||||
"ts_max": (1, ("n",)),
|
||||
"ts_min": (1, ("n",)),
|
||||
"ts_delay": (1, ("n",)),
|
||||
"ts_delta": (1, ("n",)),
|
||||
"ts_rank": (1, ("n",)),
|
||||
"ts_zscore": (1, ("n",)),
|
||||
"ts_corr": (2, ("n",)),
|
||||
"ts_cov": (2, ("n",)),
|
||||
"ts_quantile": (1, ("n", "q")),
|
||||
"decay_linear": (1, ("n",)),
|
||||
"rank": (1, ()),
|
||||
"zscore": (1, ()),
|
||||
"winsorize": (1, ("k",)), # k 可省略, 默认 3
|
||||
"power": (1, ("c",)),
|
||||
"clamp": (1, ("lo", "hi")),
|
||||
"if_else": (3, ()),
|
||||
"min": (2, ()),
|
||||
"max": (2, ()),
|
||||
"log": (1, ()),
|
||||
"abs": (1, ()),
|
||||
"sign": (1, ()),
|
||||
"sqrt": (1, ()),
|
||||
}
|
||||
TS_OPERATORS = frozenset({
|
||||
"ts_mean", "ts_std", "ts_sum", "ts_max", "ts_min", "ts_delay", "ts_delta",
|
||||
"ts_rank", "ts_zscore", "ts_corr", "ts_cov", "ts_quantile", "decay_linear",
|
||||
})
|
||||
CROSS_OPERATORS = frozenset({"rank", "zscore", "winsorize"})
|
||||
|
||||
|
||||
@dataclass
|
||||
class DslError:
|
||||
code: str
|
||||
message: str
|
||||
offset: int = 0
|
||||
detail: dict[str, Any] = field(default_factory=dict)
|
||||
|
||||
def to_dict(self) -> dict:
|
||||
return {
|
||||
"code": self.code,
|
||||
"message": self.message,
|
||||
"position": {"offset": self.offset, "line": 1},
|
||||
"detail": self.detail,
|
||||
}
|
||||
|
||||
|
||||
@dataclass
|
||||
class CompiledFormula:
|
||||
ok: bool
|
||||
errors: list[DslError] = field(default_factory=list)
|
||||
frame_transform: Any | None = None # (frame: pl.DataFrame) -> pl.DataFrame | None (缺列 None = E013)
|
||||
dependencies: frozenset[str] = frozenset() # 展开到 enriched base 列
|
||||
referenced_factors: frozenset[str] = frozenset() # 引用的注册因子 id (含 virtual, 需物化)
|
||||
warmup_bars: int = 1
|
||||
cross_sectional: bool = False
|
||||
formula_text: str = ""
|
||||
|
||||
|
||||
# ---------------------------------------------------------------- tokenizer
|
||||
|
||||
_TOKEN_RE = re.compile(
|
||||
r"\s*(?:(?P<num>\d+(?:\.\d+)?)|(?P<ident>[A-Za-z_][A-Za-z0-9_]*)|(?P<op>>=|<=|==|!=|[+\-*/><(),]))"
|
||||
)
|
||||
_KEYWORDS = frozenset({"and", "or", "not"})
|
||||
|
||||
|
||||
def _tokenize(text: str) -> tuple[list[tuple[str, Any, int]], DslError | None]:
|
||||
tokens: list[tuple[str, Any, int]] = []
|
||||
pos = 0
|
||||
while pos < len(text):
|
||||
match = _TOKEN_RE.match(text, pos)
|
||||
if match is None or match.end() == pos:
|
||||
rest = text[pos:].strip()
|
||||
if not rest:
|
||||
break
|
||||
return [], DslError("E014", f"语法错误: 无法识别的字符 '{rest[0]}'", offset=pos)
|
||||
if match.group("num") is not None:
|
||||
tokens.append(("num", float(match.group("num")), match.start("num")))
|
||||
elif match.group("ident") is not None:
|
||||
tokens.append(("ident", match.group("ident"), match.start("ident")))
|
||||
else:
|
||||
tokens.append(("op", match.group("op"), match.start("op")))
|
||||
pos = match.end()
|
||||
return tokens, None
|
||||
|
||||
|
||||
# ------------------------------------------------------------------- parser
|
||||
# AST 节点: dict(kind, value, children, offset[, _constants])
|
||||
|
||||
|
||||
class _Parser:
|
||||
_CMP = frozenset({">", ">=", "<", "<=", "==", "!="})
|
||||
|
||||
def __init__(self, tokens: list[tuple[str, Any, int]], text: str) -> None:
|
||||
self.tokens = tokens
|
||||
self.text = text
|
||||
self.index = 0
|
||||
|
||||
def _peek(self) -> tuple[str, Any, int] | None:
|
||||
return self.tokens[self.index] if self.index < len(self.tokens) else None
|
||||
|
||||
def _next(self) -> tuple[str, Any, int]:
|
||||
token = self.tokens[self.index]
|
||||
self.index += 1
|
||||
return token
|
||||
|
||||
def parse(self) -> tuple[dict | None, DslError | None]:
|
||||
if not self.tokens:
|
||||
return None, DslError("E014", "语法错误: 表达式为空", offset=0)
|
||||
node, error = self._or_expr()
|
||||
if error:
|
||||
return None, error
|
||||
if self._peek() is not None:
|
||||
_, value, offset = self._peek()
|
||||
return None, DslError("E014", f"语法错误: 多余的记号 '{value}'", offset=offset)
|
||||
return node, None
|
||||
|
||||
def _or_expr(self):
|
||||
left, error = self._and_expr()
|
||||
if error:
|
||||
return None, error
|
||||
while (token := self._peek()) and token[0] == "ident" and token[1] == "or":
|
||||
self._next()
|
||||
right, error = self._and_expr()
|
||||
if error:
|
||||
return None, error
|
||||
left = {"kind": "bin", "value": "or", "children": [left, right], "offset": token[2]}
|
||||
return left, None
|
||||
|
||||
def _and_expr(self):
|
||||
left, error = self._cmp_expr()
|
||||
if error:
|
||||
return None, error
|
||||
while (token := self._peek()) and token[0] == "ident" and token[1] == "and":
|
||||
self._next()
|
||||
right, error = self._cmp_expr()
|
||||
if error:
|
||||
return None, error
|
||||
left = {"kind": "bin", "value": "and", "children": [left, right], "offset": token[2]}
|
||||
return left, None
|
||||
|
||||
def _cmp_expr(self):
|
||||
left, error = self._add_expr()
|
||||
if error:
|
||||
return None, error
|
||||
while (token := self._peek()) and token[0] == "op" and token[1] in self._CMP:
|
||||
self._next()
|
||||
right, error = self._add_expr()
|
||||
if error:
|
||||
return None, error
|
||||
left = {"kind": "bin", "value": token[1], "children": [left, right], "offset": token[2]}
|
||||
return left, None
|
||||
|
||||
def _add_expr(self):
|
||||
left, error = self._mul_expr()
|
||||
if error:
|
||||
return None, error
|
||||
while (token := self._peek()) and token[0] == "op" and token[1] in ("+", "-"):
|
||||
self._next()
|
||||
right, error = self._mul_expr()
|
||||
if error:
|
||||
return None, error
|
||||
left = {"kind": "bin", "value": token[1], "children": [left, right], "offset": token[2]}
|
||||
return left, None
|
||||
|
||||
def _mul_expr(self):
|
||||
left, error = self._unary()
|
||||
if error:
|
||||
return None, error
|
||||
while (token := self._peek()) and token[0] == "op" and token[1] in ("*", "/"):
|
||||
self._next()
|
||||
right, error = self._unary()
|
||||
if error:
|
||||
return None, error
|
||||
left = {"kind": "bin", "value": token[1], "children": [left, right], "offset": token[2]}
|
||||
return left, None
|
||||
|
||||
def _unary(self):
|
||||
token = self._peek()
|
||||
if token and token[0] == "op" and token[1] == "-":
|
||||
self._next()
|
||||
operand, error = self._unary()
|
||||
if error:
|
||||
return None, error
|
||||
return {"kind": "unary", "value": "-", "children": [operand], "offset": token[2]}, None
|
||||
return self._primary()
|
||||
|
||||
def _primary(self):
|
||||
token = self._peek()
|
||||
if token is None:
|
||||
return None, DslError("E014", "语法错误: 表达式意外结束", offset=len(self.text))
|
||||
kind, value, offset = self._next()
|
||||
if kind == "num":
|
||||
return {"kind": "num", "value": value, "children": [], "offset": offset}, None
|
||||
if kind == "ident":
|
||||
if value in _KEYWORDS:
|
||||
return None, DslError("E014", f"语法错误: 关键字 '{value}' 不能作为操作数", offset=offset)
|
||||
nxt = self._peek()
|
||||
if nxt and nxt[0] == "op" and nxt[1] == "(":
|
||||
return self._call(value, offset)
|
||||
return {"kind": "col", "value": value, "children": [], "offset": offset}, None
|
||||
if kind == "op" and value == "(":
|
||||
inner, error = self._or_expr()
|
||||
if error:
|
||||
return None, error
|
||||
closing = self._peek()
|
||||
if not (closing and closing[0] == "op" and closing[1] == ")"):
|
||||
return None, DslError("E014", "语法错误: 缺少右括号 ')'", offset=offset)
|
||||
self._next()
|
||||
return inner, None
|
||||
return None, DslError("E014", f"语法错误: 意外的记号 '{value}'", offset=offset)
|
||||
|
||||
def _call(self, name: str, offset: int):
|
||||
self._next() # consume '('
|
||||
args: list[dict] = []
|
||||
token = self._peek()
|
||||
if not (token and token[0] == "op" and token[1] == ")"):
|
||||
while True:
|
||||
arg, error = self._or_expr()
|
||||
if error:
|
||||
return None, error
|
||||
args.append(arg)
|
||||
token = self._peek()
|
||||
if token and token[0] == "op" and token[1] == ",":
|
||||
self._next()
|
||||
continue
|
||||
break
|
||||
closing = self._peek()
|
||||
if not (closing and closing[0] == "op" and closing[1] == ")"):
|
||||
return None, DslError("E014", f"语法错误: 函数 '{name}' 缺少右括号", offset=offset)
|
||||
self._next()
|
||||
return {"kind": "call", "value": name, "children": args, "offset": offset}, None
|
||||
|
||||
|
||||
# ---------------------------------------------------------- semantic checks
|
||||
|
||||
|
||||
def _ast_depth(node: dict) -> int:
|
||||
if not node["children"]:
|
||||
return 1
|
||||
return 1 + max(_ast_depth(child) for child in node["children"])
|
||||
|
||||
|
||||
def _collect_identifiers(node: dict, found: set[str]) -> None:
|
||||
if node["kind"] == "col":
|
||||
found.add(node["value"])
|
||||
for child in node["children"]:
|
||||
_collect_identifiers(child, found)
|
||||
|
||||
|
||||
def _const_value(node: dict) -> float | None:
|
||||
if node["kind"] == "num":
|
||||
return float(node["value"])
|
||||
if node["kind"] == "unary" and node["value"] == "-" and node["children"][0]["kind"] == "num":
|
||||
return -float(node["children"][0]["value"])
|
||||
return None
|
||||
|
||||
|
||||
def _check_call(node: dict, errors: list[DslError]) -> dict[str, float]:
|
||||
"""检查函数签名与常量参数范围; 返回解析出的常量参数表。"""
|
||||
name = node["value"]
|
||||
args = node["children"]
|
||||
if name not in OPERATORS:
|
||||
errors.append(DslError("E002", f"未知函数: {name}", offset=node["offset"], detail={"name": name}))
|
||||
return {}
|
||||
n_expr, const_names = OPERATORS[name]
|
||||
has_optional_k = name == "winsorize"
|
||||
total_min, total_max = n_expr + (0 if has_optional_k else len(const_names)), n_expr + len(const_names)
|
||||
if not (total_min <= len(args) <= total_max):
|
||||
errors.append(DslError(
|
||||
"E003", f"函数 {name} 参数数量不符: 期望 {total_min}~{total_max} 个, 实际 {len(args)}",
|
||||
offset=node["offset"], detail={"name": name, "args": len(args)},
|
||||
))
|
||||
return {}
|
||||
constants: dict[str, float] = {}
|
||||
for index, const_name in enumerate(const_names):
|
||||
arg = args[n_expr + index]
|
||||
value = _const_value(arg)
|
||||
if value is None:
|
||||
errors.append(DslError(
|
||||
"E003", f"函数 {name} 的参数 {const_name} 必须是数字常量",
|
||||
offset=arg["offset"], detail={"name": name, "param": const_name},
|
||||
))
|
||||
continue
|
||||
constants[const_name] = value
|
||||
if "n" in constants:
|
||||
n_value = constants["n"]
|
||||
if n_value != int(n_value):
|
||||
errors.append(DslError("E004", "窗口参数必须是整数", offset=node["offset"], detail={"n": n_value}))
|
||||
else:
|
||||
n_int = int(n_value)
|
||||
if n_int < 0 and name in ("ts_delay", "ts_delta"):
|
||||
errors.append(DslError(
|
||||
"E005", f"负 shift: {name} 的 n 必须 ≥ 0 (负数即未来函数)",
|
||||
offset=node["offset"], detail={"n": n_int},
|
||||
))
|
||||
elif name == "ts_delay" and not (1 <= n_int <= DELAY_MAX):
|
||||
errors.append(DslError("E004", f"ts_delay 的 n 必须在 [1,{DELAY_MAX}] 内", offset=node["offset"], detail={"n": n_int}))
|
||||
elif name == "ts_delta" and not (0 <= n_int <= DELAY_MAX):
|
||||
errors.append(DslError("E004", f"ts_delta 的 n 必须在 [0,{DELAY_MAX}] 内", offset=node["offset"], detail={"n": n_int}))
|
||||
elif name not in ("ts_delay", "ts_delta") and not (WINDOW_MIN <= n_int <= WINDOW_MAX):
|
||||
errors.append(DslError(
|
||||
"E004", f"窗口 n 必须在 [{WINDOW_MIN},{WINDOW_MAX}] 内", offset=node["offset"], detail={"n": n_int},
|
||||
))
|
||||
if "q" in constants and not (0.0 < constants["q"] < 1.0):
|
||||
errors.append(DslError("E004", "ts_quantile 的 q 必须在 (0,1) 开区间内", offset=node["offset"], detail={"q": constants["q"]}))
|
||||
if "c" in constants and abs(constants["c"]) > POWER_ABS_MAX:
|
||||
errors.append(DslError("E010", f"power 指数 |c| ≤ {POWER_ABS_MAX}", offset=node["offset"], detail={"c": constants["c"]}))
|
||||
if "k" in constants and not (WINSORIZE_K_RANGE[0] <= constants["k"] <= WINSORIZE_K_RANGE[1]):
|
||||
errors.append(DslError("E011", "winsorize 的 k 必须在 [1,6] 内", offset=node["offset"], detail={"k": constants["k"]}))
|
||||
if "lo" in constants and "hi" in constants and constants["lo"] > constants["hi"]:
|
||||
errors.append(DslError("E003", "clamp 的 lo 不能大于 hi", offset=node["offset"]))
|
||||
return constants
|
||||
|
||||
|
||||
def _semantic_walk(node: dict, errors: list[DslError], constants_by_call: dict[int, dict]) -> None:
|
||||
if node["kind"] == "call":
|
||||
constants_by_call[id(node)] = _check_call(node, errors)
|
||||
for child in node["children"]:
|
||||
_semantic_walk(child, errors, constants_by_call)
|
||||
return
|
||||
if node["kind"] == "bin" and node["value"] == "/":
|
||||
right = node["children"][1]
|
||||
if _const_value(right) == 0:
|
||||
errors.append(DslError("E008", "静态除零: 分母为常量 0", offset=right["offset"]))
|
||||
for child in node["children"]:
|
||||
_semantic_walk(child, errors, constants_by_call)
|
||||
|
||||
|
||||
# ------------------------------------------------------------- code generation
|
||||
|
||||
_CMP_METHOD = {">": "gt", ">=": "ge", "<": "lt", "<=": "le", "==": "eq", "!=": "ne"}
|
||||
|
||||
|
||||
def _safe_div(numerator: pl.Expr, denominator: pl.Expr) -> pl.Expr:
|
||||
return (
|
||||
pl.when(denominator.is_not_null() & (denominator != 0))
|
||||
.then(numerator / denominator)
|
||||
.otherwise(None)
|
||||
)
|
||||
|
||||
|
||||
def _rolling_apply(inner: pl.Expr, op: str, n: int, extra: dict[str, float]) -> pl.Expr:
|
||||
"""对无 over 的内层序列应用窗口逻辑; 返回值同样不挂 over。"""
|
||||
if op == "ts_mean":
|
||||
return inner.rolling_mean(n, min_samples=n)
|
||||
if op == "ts_std":
|
||||
return inner.rolling_std(n, min_samples=n)
|
||||
if op == "ts_sum":
|
||||
return inner.rolling_sum(n, min_samples=n)
|
||||
if op == "ts_max":
|
||||
return inner.rolling_max(n, min_samples=n)
|
||||
if op == "ts_min":
|
||||
return inner.rolling_min(n, min_samples=n)
|
||||
if op == "ts_delay":
|
||||
return inner.shift(n)
|
||||
if op == "ts_delta":
|
||||
return inner - inner.shift(n)
|
||||
if op == "ts_rank":
|
||||
return inner.rolling_rank(n, min_samples=n)
|
||||
if op == "ts_zscore":
|
||||
mean = inner.rolling_mean(n, min_samples=n)
|
||||
std = inner.rolling_std(n, min_samples=n)
|
||||
return pl.when(std > 0).then((inner - mean) / std).otherwise(None)
|
||||
if op == "ts_quantile":
|
||||
return inner.rolling_quantile(extra.get("q", 0.5), window_size=n, min_samples=n)
|
||||
if op == "decay_linear":
|
||||
# 近端权重大: 权重 n, n-1, ..., 1, 总权 n(n+1)/2
|
||||
weighted = None
|
||||
for i in range(n):
|
||||
term = (n - i) * inner.shift(i)
|
||||
weighted = term if weighted is None else weighted + term
|
||||
assert weighted is not None
|
||||
return _safe_div(weighted, pl.lit(float(n * (n + 1) / 2)))
|
||||
raise AssertionError(op)
|
||||
|
||||
|
||||
def _compile_node(node: dict) -> tuple[pl.Expr | None, bool, bool]:
|
||||
"""返回 (expr, needs_symbol_window, is_bool)。
|
||||
|
||||
needs_symbol_window=True 表示该子树含 ts 窗口逻辑但尚未挂 over;
|
||||
由非时序上下文的调用方挂 over("symbol"), 时序上下文继续向内传递。
|
||||
"""
|
||||
kind = node["kind"]
|
||||
if kind == "num":
|
||||
return pl.lit(node["value"]), False, False
|
||||
if kind == "col":
|
||||
# 基准列/base 因子/虚拟因子统一以列引用; 虚拟因子由调用方物化 (运行期缺列 fail-closed)
|
||||
return pl.col(node["value"]), False, False
|
||||
if kind == "unary":
|
||||
operand, needs_window, _ = _compile_node(node["children"][0])
|
||||
if operand is None:
|
||||
return None, False, False
|
||||
return -operand, needs_window, False
|
||||
if kind == "bin":
|
||||
op = node["value"]
|
||||
left, left_window, _ = _compile_node(node["children"][0])
|
||||
right, right_window, _ = _compile_node(node["children"][1])
|
||||
if left is None or right is None:
|
||||
return None, False, False
|
||||
if left_window:
|
||||
left = left.over("symbol")
|
||||
if right_window:
|
||||
right = right.over("symbol")
|
||||
if op == "+":
|
||||
return left + right, False, False
|
||||
if op == "-":
|
||||
return left - right, False, False
|
||||
if op == "*":
|
||||
return left * right, False, False
|
||||
if op == "/":
|
||||
return _safe_div(left, right), False, False
|
||||
if op in _CMP_METHOD:
|
||||
return getattr(left, _CMP_METHOD[op])(right), False, True
|
||||
if op == "and":
|
||||
return left & right, False, True
|
||||
if op == "or":
|
||||
return left | right, False, True
|
||||
return None, False, False
|
||||
if kind == "call":
|
||||
return _compile_call(node)
|
||||
return None, False, False
|
||||
|
||||
|
||||
def _compile_call(node: dict) -> tuple[pl.Expr | None, bool, bool]:
|
||||
name = node["value"]
|
||||
children = node["children"]
|
||||
constants: dict[str, float] = node.get("_constants", {})
|
||||
n_expr, _ = OPERATORS[name]
|
||||
|
||||
if name in TS_OPERATORS:
|
||||
inner, _, _ = _compile_node(children[0])
|
||||
if inner is None:
|
||||
return None, False, False
|
||||
if name in ("ts_corr", "ts_cov"):
|
||||
second, _, _ = _compile_node(children[1])
|
||||
if second is None:
|
||||
return None, False, False
|
||||
n = int(constants.get("n", 0))
|
||||
expr = (
|
||||
pl.rolling_corr(inner, second, window_size=n)
|
||||
if name == "ts_corr"
|
||||
else pl.rolling_cov(inner, second, window_size=n)
|
||||
)
|
||||
return expr, True, False
|
||||
expr = _rolling_apply(inner, name, int(constants.get("n", 0)), constants)
|
||||
return expr, True, False
|
||||
|
||||
if name in CROSS_OPERATORS:
|
||||
inner, inner_window, _ = _compile_node(children[0])
|
||||
if inner is None:
|
||||
return None, False, False
|
||||
if inner_window:
|
||||
inner = inner.over("symbol")
|
||||
if name == "rank":
|
||||
count = inner.count().over("date")
|
||||
return inner.rank(method="average").over("date") / count, False, False
|
||||
if name == "zscore":
|
||||
mean = inner.mean().over("date")
|
||||
std = inner.std().over("date")
|
||||
return pl.when(std > 0).then((inner - mean) / std).otherwise(None), False, False
|
||||
k = constants.get("k", 3.0)
|
||||
mean = inner.mean().over("date")
|
||||
std = inner.std().over("date")
|
||||
return inner.clip(mean - k * std, mean + k * std), False, False
|
||||
|
||||
if name == "if_else":
|
||||
cond, cond_window, _ = _compile_node(children[0])
|
||||
then_expr, then_window, _ = _compile_node(children[1])
|
||||
else_expr, else_window, _ = _compile_node(children[2])
|
||||
if cond is None or then_expr is None or else_expr is None:
|
||||
return None, False, False
|
||||
if cond_window:
|
||||
cond = cond.over("symbol")
|
||||
if then_window:
|
||||
then_expr = then_expr.over("symbol")
|
||||
if else_window:
|
||||
else_expr = else_expr.over("symbol")
|
||||
return pl.when(cond).then(then_expr).otherwise(else_expr), False, False
|
||||
|
||||
args: list[pl.Expr | None] = []
|
||||
arg_windows: list[bool] = []
|
||||
for index in range(n_expr):
|
||||
arg, arg_window, _ = _compile_node(children[index])
|
||||
args.append(arg)
|
||||
arg_windows.append(arg_window)
|
||||
if any(arg is None for arg in args):
|
||||
return None, False, False
|
||||
resolved: list[pl.Expr] = []
|
||||
for arg, arg_window in zip(args, arg_windows, strict=True):
|
||||
resolved.append(arg.over("symbol") if arg_window else arg)
|
||||
first = resolved[0]
|
||||
if name == "log":
|
||||
return pl.when(first > 0).then(first.log()).otherwise(None), False, False
|
||||
if name == "abs":
|
||||
return first.abs(), False, False
|
||||
if name == "sign":
|
||||
return first.sign(), False, False
|
||||
if name == "sqrt":
|
||||
return pl.when(first >= 0).then(first.sqrt()).otherwise(None), False, False
|
||||
if name == "power":
|
||||
return first.pow(constants.get("c", 1.0)), False, False
|
||||
if name == "clamp":
|
||||
return first.clip(constants.get("lo"), constants.get("hi")), False, False
|
||||
if name == "min":
|
||||
return pl.min_horizontal(*resolved), False, False
|
||||
if name == "max":
|
||||
return pl.max_horizontal(*resolved), False, False
|
||||
return None, False, False
|
||||
|
||||
|
||||
def compile_formula(text: str) -> CompiledFormula:
|
||||
"""编译公式文本; 永不抛异常, 失败以 errors 表达 (fail-closed)。"""
|
||||
if not isinstance(text, str) or not text.strip():
|
||||
return CompiledFormula(ok=False, errors=[DslError("E014", "语法错误: 表达式为空")], formula_text=text)
|
||||
|
||||
tokens, tokenize_error = _tokenize(text)
|
||||
errors: list[DslError] = [tokenize_error] if tokenize_error else []
|
||||
if len(tokens) > MAX_TOKENS:
|
||||
errors.append(DslError("E007", f"规模超限: token 数 {len(tokens)} > {MAX_TOKENS}"))
|
||||
if errors:
|
||||
return CompiledFormula(ok=False, errors=errors, formula_text=text)
|
||||
|
||||
ast, parse_error = _Parser(tokens, text).parse()
|
||||
if parse_error:
|
||||
return CompiledFormula(ok=False, errors=[parse_error], formula_text=text)
|
||||
|
||||
if _ast_depth(ast) > MAX_AST_DEPTH:
|
||||
errors.append(DslError("E006", f"嵌套深度超限: AST 深度 {_ast_depth(ast)} > {MAX_AST_DEPTH}"))
|
||||
|
||||
identifiers: set[str] = set()
|
||||
_collect_identifiers(ast, identifiers)
|
||||
if not identifiers:
|
||||
errors.append(DslError("E016", "常量表达式: 公式必须引用至少一个数据列或因子"))
|
||||
|
||||
for name in sorted(identifiers):
|
||||
if name not in BASE_COLUMNS and get_factor(name) is None:
|
||||
errors.append(DslError("E001", f"未知标识符: {name}", detail={"name": name}))
|
||||
|
||||
constants_by_call: dict[int, dict] = {}
|
||||
_semantic_walk(ast, errors, constants_by_call)
|
||||
|
||||
dependencies: set[str] = set()
|
||||
referenced_factors: set[str] = set()
|
||||
warmup = 1
|
||||
cross_sectional = False
|
||||
for name in identifiers:
|
||||
if name in BASE_COLUMNS:
|
||||
dependencies.add(name)
|
||||
continue
|
||||
spec = get_factor(name)
|
||||
if spec is None:
|
||||
continue
|
||||
referenced_factors.add(name)
|
||||
dependencies.update(factor_dependencies([name]))
|
||||
warmup = max(warmup, spec.warmup_bars)
|
||||
|
||||
for node_constants in constants_by_call.values():
|
||||
n_value = node_constants.get("n")
|
||||
if n_value is not None and n_value == int(n_value) and int(n_value) > 0:
|
||||
warmup = max(warmup, int(n_value) + 1)
|
||||
|
||||
def _find_cross(node: dict) -> None:
|
||||
nonlocal cross_sectional
|
||||
if node["kind"] == "call" and node["value"] in CROSS_OPERATORS:
|
||||
cross_sectional = True
|
||||
for child in node["children"]:
|
||||
_find_cross(child)
|
||||
|
||||
_find_cross(ast)
|
||||
|
||||
if errors:
|
||||
return CompiledFormula(
|
||||
ok=False, errors=errors, dependencies=frozenset(dependencies),
|
||||
referenced_factors=frozenset(referenced_factors),
|
||||
warmup_bars=warmup, cross_sectional=cross_sectional, formula_text=text,
|
||||
)
|
||||
|
||||
# 挂常量表必须在任何 deepcopy 之前 (deepcopy 携带 _constants; 事后按 id() 重挂会失联)
|
||||
def _attach(node: dict) -> None:
|
||||
if node["kind"] == "call":
|
||||
node["_constants"] = constants_by_call.get(id(node), {})
|
||||
for child in node["children"]:
|
||||
_attach(child)
|
||||
|
||||
_attach(ast)
|
||||
|
||||
# 阶段一: 校验并拒绝"截面算子嵌在时序窗口内" (无法单层 over 表达)
|
||||
def _contains_cross(node: dict) -> bool:
|
||||
if node["kind"] == "call" and node["value"] in CROSS_OPERATORS:
|
||||
return True
|
||||
return any(_contains_cross(child) for child in node["children"])
|
||||
|
||||
def _reject_cross_in_ts(node: dict) -> None:
|
||||
if node["kind"] == "call" and node["value"] in TS_OPERATORS:
|
||||
for child in node["children"]:
|
||||
if _contains_cross(child):
|
||||
errors.append(DslError(
|
||||
"E009",
|
||||
f"截面算子不能嵌在时序窗口内: {node['value']}(...) 的参数含 rank/zscore/winsorize",
|
||||
offset=node["offset"],
|
||||
))
|
||||
return
|
||||
for child in node["children"]:
|
||||
_reject_cross_in_ts(child)
|
||||
|
||||
_reject_cross_in_ts(ast)
|
||||
if errors:
|
||||
return CompiledFormula(
|
||||
ok=False, errors=errors, dependencies=frozenset(dependencies),
|
||||
referenced_factors=frozenset(referenced_factors),
|
||||
warmup_bars=warmup, cross_sectional=cross_sectional, formula_text=text,
|
||||
)
|
||||
|
||||
# 阶段二: 提取截面算子的含窗口子树为临时列 (Polars 嵌套窗口会静默全 null)
|
||||
# worklist 逐层下钻; temps 后进先出反转即依赖顺序 (深层先算)。
|
||||
def _needs_symbol_window(node: dict) -> bool:
|
||||
kind = node["kind"]
|
||||
if kind in ("num", "col"):
|
||||
return False
|
||||
if kind == "unary":
|
||||
return _needs_symbol_window(node["children"][0])
|
||||
if node["kind"] == "call" and node["value"] in TS_OPERATORS:
|
||||
return True
|
||||
return any(_needs_symbol_window(child) for child in node["children"])
|
||||
|
||||
def _has_any_over(node: dict) -> bool:
|
||||
# 含时序窗口 或 含截面算子(编译后自带 over("date")) 的子树都不能直接进截面上下文
|
||||
return _needs_symbol_window(node) or _contains_cross(node)
|
||||
|
||||
temp_roots: list[dict] = []
|
||||
pending: list[dict] = [ast]
|
||||
while pending:
|
||||
current = pending.pop(0)
|
||||
if current.get("kind") == "call" and current.get("value") in CROSS_OPERATORS:
|
||||
operand = current["children"][0]
|
||||
if _has_any_over(operand):
|
||||
alias = f"__tsfx_{len(temp_roots)}__"
|
||||
current["children"][0] = {"kind": "col", "value": alias, "children": [], "offset": operand["offset"]}
|
||||
temp_roots.append({"alias": alias, "root": copy.deepcopy(operand)})
|
||||
pending.append(temp_roots[-1]["root"])
|
||||
continue # 操作数已替换为临时列, 不再下钻原子树
|
||||
pending.extend(current.get("children", []))
|
||||
|
||||
# 阶段三: 编译最终表达式与临时列表达式 (按依赖顺序: 深层在前)
|
||||
# _constants 已在 deepcopy 前挂载并被复制携带, 不得按 id() 重挂 (复制后 id 失联)
|
||||
temp_exprs: list[pl.Expr] = []
|
||||
for item in reversed(temp_roots):
|
||||
root = copy.deepcopy(item["root"])
|
||||
expr, needs_window, _ = _compile_node(root)
|
||||
if expr is None:
|
||||
errors.append(DslError("E009", f"无法编译临时列: {item['alias']}"))
|
||||
continue
|
||||
if needs_window:
|
||||
expr = expr.over("symbol")
|
||||
temp_exprs.append(expr.alias(item["alias"]))
|
||||
|
||||
final_ast = copy.deepcopy(ast)
|
||||
compiled, needs_window, is_bool = _compile_node(final_ast)
|
||||
if compiled is None or errors:
|
||||
return CompiledFormula(
|
||||
ok=False,
|
||||
errors=errors or [DslError("E009", "产出类型非法: 无法编译为数值表达式")],
|
||||
dependencies=frozenset(dependencies),
|
||||
referenced_factors=frozenset(referenced_factors),
|
||||
warmup_bars=warmup, cross_sectional=cross_sectional, formula_text=text,
|
||||
)
|
||||
if needs_window:
|
||||
compiled = compiled.over("symbol")
|
||||
if is_bool:
|
||||
compiled = compiled.cast(pl.Float64)
|
||||
|
||||
# 运行期帧变换: 检查全部引用列 (基准依赖 + 引用因子) 存在, 否则 None (E013 fail-closed)
|
||||
required_columns = set(dependencies) | set(referenced_factors)
|
||||
staged_exprs = temp_exprs # 依赖顺序已排
|
||||
|
||||
def frame_transform(frame: pl.DataFrame) -> pl.DataFrame | None:
|
||||
if not required_columns.issubset(set(frame.columns)):
|
||||
return None
|
||||
result = frame
|
||||
if staged_exprs:
|
||||
result = result.with_columns(staged_exprs)
|
||||
return result.with_columns(compiled.alias(FACTOR_COLUMN))
|
||||
|
||||
return CompiledFormula(
|
||||
ok=True,
|
||||
errors=[],
|
||||
frame_transform=frame_transform,
|
||||
dependencies=frozenset(dependencies),
|
||||
referenced_factors=frozenset(referenced_factors),
|
||||
warmup_bars=warmup,
|
||||
cross_sectional=cross_sectional,
|
||||
formula_text=text,
|
||||
)
|
||||
|
||||
|
||||
@lru_cache(maxsize=256)
|
||||
def compile_formula_cached(text: str) -> CompiledFormula:
|
||||
"""带 LRU 缓存的编译入口 (公式文本 → 编译产物, 设计文档 §3.3)。
|
||||
|
||||
CompiledFormula 为不可变值对象 (frame_transform 闭包只读), 缓存共享安全。
|
||||
"""
|
||||
return compile_formula(text)
|
||||
@@ -0,0 +1,383 @@
|
||||
"""因子注册表 (L-REG) — 因子元数据的单一权威来源。
|
||||
|
||||
P1 收口范围: 目录元数据 (id/label/group/公式)、虚拟因子依赖声明、评分预热窗口。
|
||||
三处历史清单在此合一:
|
||||
- backtest/factor.py FACTOR_COLUMNS (由 factor_columns_view() 生成兼容别名)
|
||||
- strategy/scoring.py VIRTUAL_SCORING_DEPENDENCIES (由 virtual_dependencies() 生成)
|
||||
- strategy/scoring.py _ROLLING_SCORING_WARMUP (由 scoring_warmups() 生成)
|
||||
|
||||
P1 边界 (诚实声明):
|
||||
- scoring_value_expr 的表达式分发仍留在 scoring.py, 注册表不含计算逻辑;
|
||||
复合/自定义因子 (composite/custom) 与 DSL 在 P2/P3 接入后再收口。
|
||||
- unit 字段 P1 统一 "none": 单位口径涉及金融数据契约 (CONTRIBUTING §3),
|
||||
未经逐因子核对禁止猜测填充; 前端 P1 也不按 unit 格式化。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Literal
|
||||
|
||||
Kind = Literal["base", "virtual", "composite", "custom"]
|
||||
Direction = Literal["high", "low", "none"]
|
||||
Unit = Literal["ratio", "pct", "score", "count", "days", "currency", "none"]
|
||||
PitSource = Literal["financial_announce", "share_capital_announce", "none"]
|
||||
Stability = Literal["stable", "experimental", "deprecated"]
|
||||
|
||||
_ALL_ASSETS = frozenset({"stock", "etf"})
|
||||
_STOCK_ONLY = frozenset({"stock"})
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class FactorSpec:
|
||||
id: str
|
||||
label: str
|
||||
group: str
|
||||
formula_text: str
|
||||
kind: Kind = "base"
|
||||
version: int = 1
|
||||
# base: 空集合 = 已物化列自身; virtual: 展开到 enriched base 列
|
||||
dependencies: frozenset[str] = field(default_factory=frozenset)
|
||||
direction: Direction = "none" # P1 不预填: 方向以最近检验 IC 符号为准 (见平台方案 §3.6)
|
||||
unit: Unit = "none"
|
||||
warmup_bars: int = 1
|
||||
pit: bool = False
|
||||
pit_source: PitSource = "none"
|
||||
asset_types: frozenset[str] = _ALL_ASSETS
|
||||
incremental_safe: bool = True
|
||||
scale_free: bool = True
|
||||
null_policy: Literal["keep", "drop_row"] = "keep"
|
||||
stability: Stability = "stable"
|
||||
tags: tuple[str, ...] = ()
|
||||
# composite 专用: ((成员 id, 权重), ...); 其余类型为空
|
||||
components: tuple[tuple[str, float], ...] = ()
|
||||
|
||||
def column_view(self) -> dict:
|
||||
"""历史 FACTOR_COLUMNS 条目视图 (键与顺序兼容)。"""
|
||||
return {"id": self.id, "label": self.label, "group": self.group, "desc": self.formula_text}
|
||||
|
||||
|
||||
def _base(fid: str, label: str, group: str, desc: str, **overrides) -> FactorSpec:
|
||||
return FactorSpec(id=fid, label=label, group=group, formula_text=desc, kind="base", **overrides)
|
||||
|
||||
|
||||
def _virtual(fid: str, label: str, group: str, desc: str, deps: frozenset[str], **overrides) -> FactorSpec:
|
||||
return FactorSpec(
|
||||
id=fid, label=label, group=group, formula_text=desc,
|
||||
kind="virtual", dependencies=deps, **overrides,
|
||||
)
|
||||
|
||||
|
||||
def _financial(fid: str, label: str, desc: str) -> FactorSpec:
|
||||
return FactorSpec(
|
||||
id=fid, label=label, group="财务", formula_text=desc,
|
||||
kind="base", pit=True, pit_source="financial_announce", asset_types=_STOCK_ONLY,
|
||||
)
|
||||
|
||||
|
||||
# 顺序即历史 FACTOR_COLUMNS 顺序 (mining_schedule 取前 48 个, 不得重排)。
|
||||
_CATALOG: tuple[FactorSpec, ...] = (
|
||||
# --- 动量 ---
|
||||
_base("momentum_5d", "5日动量", "动量", "5个交易日累计收益率"),
|
||||
_base("momentum_10d", "10日动量", "动量", "10个交易日累计收益率"),
|
||||
_base("momentum_20d", "20日动量", "动量", "20个交易日累计收益率"),
|
||||
_base("momentum_30d", "30日动量", "动量", "30个交易日累计收益率"),
|
||||
_base("momentum_60d", "60日动量", "动量", "60个交易日累计收益率"),
|
||||
_base("change_pct", "日涨跌幅", "动量", "当日收盘相对前收盘的收益率"),
|
||||
# --- 均线偏离 (虚拟) ---
|
||||
*(
|
||||
_virtual(
|
||||
f"ma{period}_bias", f"MA{period}乖离", "均线偏离", f"收盘价 / MA{period} - 1",
|
||||
deps=frozenset({"close", f"ma{period}"}),
|
||||
)
|
||||
for period in (5, 10, 20, 30, 60)
|
||||
),
|
||||
*(
|
||||
_virtual(
|
||||
f"ema{period}_bias", f"EMA{period}乖离", "均线偏离", f"收盘价 / EMA{period} - 1",
|
||||
deps=frozenset({"close", f"ema{period}"}),
|
||||
)
|
||||
for period in (5, 10, 20, 30, 60)
|
||||
),
|
||||
# --- 超买超卖 ---
|
||||
_base("rsi_6", "RSI(6)", "超买超卖", "6日相对强弱指标"),
|
||||
_base("rsi_14", "RSI(14)", "超买超卖", "14日相对强弱指标"),
|
||||
_base("rsi_24", "RSI(24)", "超买超卖", "24日相对强弱指标"),
|
||||
# --- 趋势 ---
|
||||
_base(
|
||||
"macd_hist", "MACD柱(原值)", "趋势",
|
||||
"兼容历史研究; 跨股票比较建议优先使用MACD柱强度",
|
||||
scale_free=False,
|
||||
),
|
||||
_virtual("macd_dif_pct", "MACD DIF强度", "趋势", "MACD DIF / 收盘价", deps=frozenset({"close", "macd_dif"})),
|
||||
_virtual("macd_dea_pct", "MACD DEA强度", "趋势", "MACD DEA / 收盘价", deps=frozenset({"close", "macd_dea"})),
|
||||
_virtual("macd_hist_pct", "MACD柱强度", "趋势", "MACD柱 / 收盘价, 消除股价尺度影响", deps=frozenset({"close", "macd_hist"})),
|
||||
_base("kdj_k", "KDJ-K", "趋势", "KDJ指标K值"),
|
||||
_base("kdj_d", "KDJ-D", "趋势", "KDJ指标D值"),
|
||||
_base("kdj_j", "KDJ-J", "趋势", "KDJ指标J值"),
|
||||
_virtual(
|
||||
"boll_position", "布林位置", "趋势", "收盘价在布林带下轨到上轨之间的位置",
|
||||
deps=frozenset({"close", "boll_upper", "boll_lower"}),
|
||||
),
|
||||
# --- 波动率 ---
|
||||
_base("annual_vol_20d", "20日波动率", "波动率", "20日收益率年化标准差"),
|
||||
_base("atr_14", "ATR(14)原值", "波动率", "兼容历史研究; 跨股票比较建议优先使用ATR相对波动", scale_free=False),
|
||||
_virtual("atr_pct", "ATR相对波动", "波动率", "ATR(14) / 收盘价", deps=frozenset({"close", "atr_14"})),
|
||||
_base("amplitude", "日振幅", "波动率", "当日高低价差 / 前收盘价"),
|
||||
_virtual(
|
||||
"boll_width", "布林带宽", "波动率", "布林带上下轨宽度 / MA20",
|
||||
deps=frozenset({"ma20", "boll_upper", "boll_lower"}),
|
||||
),
|
||||
# --- 量价 ---
|
||||
_base("vol_ratio_5d", "5日量比", "量价", "当日成交量 / 前5日平均成交量"),
|
||||
_virtual(
|
||||
"vol_ratio_10d", "10日量比", "量价", "当日成交量 / 前10日平均成交量",
|
||||
deps=frozenset({"volume"}), warmup_bars=11,
|
||||
),
|
||||
_virtual(
|
||||
"vol_trend_5_10", "成交量趋势", "量价", "5日平均成交量 / 10日平均成交量 - 1",
|
||||
deps=frozenset({"vol_ma5", "vol_ma10"}),
|
||||
),
|
||||
_base("turnover_rate", "换手率", "量价", "使用历史时点流通股本计算的当日换手率"),
|
||||
_virtual(
|
||||
"turnover_ratio_5d", "换手率放大", "量价", "当日换手率 / 前5日平均换手率 - 1",
|
||||
deps=frozenset({"turnover_rate"}), warmup_bars=6,
|
||||
),
|
||||
_virtual(
|
||||
"log_amount", "成交额对数", "量价", "ln(成交额 + 1), 降低极端规模影响",
|
||||
deps=frozenset({"amount"}), scale_free=False,
|
||||
),
|
||||
_virtual(
|
||||
"amount_ratio_5d", "成交额放大", "量价", "当日成交额 / 前5日平均成交额 - 1",
|
||||
deps=frozenset({"amount"}), warmup_bars=6,
|
||||
),
|
||||
# --- 价格位置 ---
|
||||
_virtual("gap_return", "开盘跳空", "价格位置", "开盘价 / 前收盘价 - 1", deps=frozenset({"open", "prev_close"})),
|
||||
_virtual("intraday_return", "日内收益", "价格位置", "收盘价 / 开盘价 - 1", deps=frozenset({"open", "close"})),
|
||||
_virtual(
|
||||
"close_position", "收盘位置", "价格位置", "收盘价在当日最低价到最高价之间的位置",
|
||||
deps=frozenset({"high", "low", "close"}),
|
||||
),
|
||||
_virtual(
|
||||
"distance_to_high_60d", "距60日高点", "价格位置", "收盘价 / 60日最高收盘价 - 1",
|
||||
deps=frozenset({"close", "high_60d"}),
|
||||
),
|
||||
_virtual(
|
||||
"distance_from_low_60d", "距60日低点", "价格位置", "收盘价 / 60日最低收盘价 - 1",
|
||||
deps=frozenset({"close", "low_60d"}),
|
||||
),
|
||||
_virtual(
|
||||
"vwap_bias", "VWAP乖离", "价格位置", "收盘价 / 当日成交均价 - 1, 成交均价 = 成交额 / (成交量x100)",
|
||||
deps=frozenset({"close", "volume", "amount"}),
|
||||
),
|
||||
# --- 收益形态 (虚拟, 滚动窗口) ---
|
||||
_virtual(
|
||||
"max_ret_20d", "20日最大单日涨幅", "收益形态", "近20个交易日单日涨幅最大值(彩票效应, 高值代表博彩型特征强)",
|
||||
deps=frozenset({"close"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"ret_skew_20d", "20日收益偏度", "收益形态", "近20个交易日日收益偏度, 高值代表右偏(偶发大涨)",
|
||||
deps=frozenset({"close"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"up_days_20d", "20日上涨天数", "收益形态", "近20个交易日中上涨天数(0~20)",
|
||||
deps=frozenset({"close"}), warmup_bars=21,
|
||||
),
|
||||
# --- 流动性 (虚拟) ---
|
||||
_virtual(
|
||||
"amihud_20d", "20日Amihud非流动性", "流动性", "近20日平均 |日涨跌幅| / 成交额(亿元), 高值代表流动性差",
|
||||
deps=frozenset({"close", "amount"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"turnover_z_60d", "换手率60日z分", "流动性", "(当日换手率 - 前60日均值) / 前60日标准差, 衡量换手异动",
|
||||
deps=frozenset({"turnover_rate"}), warmup_bars=61,
|
||||
),
|
||||
# --- 量价 (续) ---
|
||||
_virtual(
|
||||
"vol_price_corr_20d", "20日量价相关", "量价", "近20个交易日日涨跌幅与成交量的相关系数, 高值代表量价同向",
|
||||
deps=frozenset({"close", "volume"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"vol_trend_5_60", "量能趋势(5/60)", "量价", "5日平均成交量 / 60日平均成交量 - 1",
|
||||
deps=frozenset({"volume"}), warmup_bars=60,
|
||||
),
|
||||
# --- 涨停基因 (虚拟) ---
|
||||
_virtual(
|
||||
"limit_up_count_20d", "涨停基因(20日)", "涨停基因", "近20个交易日涨停次数",
|
||||
deps=frozenset({"consecutive_limit_ups"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"limit_up_count_60d", "涨停基因(60日)", "涨停基因", "近60个交易日涨停次数",
|
||||
deps=frozenset({"consecutive_limit_ups"}), warmup_bars=61,
|
||||
),
|
||||
# --- 财务 (点时, 仅股票) ---
|
||||
_financial("pb_latest", "市净率(最新公告)", "收盘价 / 最新已公告每股净资产; 无财务数据或公告前为空"),
|
||||
_financial("roe_latest", "ROE(最新公告)", "最新已公告净资产收益率(%); 无财务数据或公告前为空"),
|
||||
_financial("gross_margin_latest", "毛利率(最新公告)", "最新已公告销售毛利率(%)"),
|
||||
_financial("net_margin_latest", "净利率(最新公告)", "最新已公告销售净利率(%)"),
|
||||
_financial("revenue_yoy_latest", "营收增速(最新公告)", "最新已公告营业收入同比(%)"),
|
||||
_financial("net_income_yoy_latest", "净利增速(最新公告)", "最新已公告归母净利润同比(%)"),
|
||||
_financial("debt_ratio_latest", "资产负债率(最新公告)", "最新已公告资产负债率(%)"),
|
||||
# --- 扩充批次 (2026-09-05): 规模/收益分解/长窗口/下行风险/量能潮/换手水平 ---
|
||||
_virtual(
|
||||
"log_float_mv", "流通市值对数", "规模",
|
||||
"ln(收盘价 x 当日成交量 / 换手率), 由换手率反推流通股本, 高值代表大盘",
|
||||
deps=frozenset({"close", "volume", "turnover_rate"}), scale_free=False,
|
||||
),
|
||||
_virtual(
|
||||
"momentum_120d", "120日动量", "动量",
|
||||
"120个交易日累计收益率 (中期动量, 与短窗口互补)",
|
||||
deps=frozenset({"close"}), warmup_bars=121,
|
||||
),
|
||||
_virtual(
|
||||
"mom_accel_20_60", "动量加速度", "动量",
|
||||
"20日动量 - 60日动量, 衡量近期动量相对中期是否增强",
|
||||
deps=frozenset({"momentum_20d", "momentum_60d"}),
|
||||
),
|
||||
_virtual(
|
||||
"rsi_14_delta_5d", "RSI五日变化", "超买超卖",
|
||||
"RSI(14) - 5日前的RSI(14), 衡量强弱指标的边际变化",
|
||||
deps=frozenset({"rsi_14"}), warmup_bars=6,
|
||||
),
|
||||
_virtual(
|
||||
"overnight_ret_20d", "20日隔夜收益", "收益形态",
|
||||
"近20日累计隔夜收益(开盘价/前收盘-1求和), A股隔夜与日内收益的定价机制不同",
|
||||
deps=frozenset({"open", "prev_close"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"intraday_ret_20d", "20日日内收益", "收益形态",
|
||||
"近20日累计日内收益(收盘价/开盘价-1求和), 与隔夜收益构成收益分解",
|
||||
deps=frozenset({"open", "close"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"downside_vol_20d", "20日下行波动", "波动率",
|
||||
"sqrt(近20日 min(日收益,0)^2 均值), 只度量下跌侧风险",
|
||||
deps=frozenset({"close"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"vol_regime_5_60", "波动率状态(5/60)", "波动率",
|
||||
"5日收益标准差 / 60日收益标准差, 高值代表波动骤然放大",
|
||||
deps=frozenset({"close"}), warmup_bars=61,
|
||||
),
|
||||
_virtual(
|
||||
"amplitude_trend_20_60", "振幅趋势(20/60)", "波动率",
|
||||
"20日平均振幅 / 60日平均振幅 - 1",
|
||||
deps=frozenset({"amplitude"}), warmup_bars=61,
|
||||
),
|
||||
_virtual(
|
||||
"obv_trend_20d", "20日量能潮", "量价",
|
||||
"近20日 sign(日收益)x成交量 之和 / (20日均量x20), 有界[-1,1], 净买入方向的一致性",
|
||||
deps=frozenset({"close", "volume"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"amount_mean_20d", "20日均成交额(亿)", "量价",
|
||||
"近20日平均成交额(亿元), 规模/流动性水平量",
|
||||
deps=frozenset({"amount"}), warmup_bars=21, scale_free=False,
|
||||
),
|
||||
_virtual(
|
||||
"turnover_mean_20d", "20日均换手", "流动性",
|
||||
"近20日平均换手率, A股经典低换手溢价因子",
|
||||
deps=frozenset({"turnover_rate"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"turnover_std_20d", "20日换手波动", "流动性",
|
||||
"近20日换手率标准差 / 均值 (变异系数), 衡量交易活跃的稳定性",
|
||||
deps=frozenset({"turnover_rate"}), warmup_bars=21,
|
||||
),
|
||||
_virtual(
|
||||
"position_240d", "一年价格位置", "价格位置",
|
||||
"收盘价在近240个交易日最低价到最高价之间的位置 (0~1)",
|
||||
deps=frozenset({"close"}), warmup_bars=241,
|
||||
),
|
||||
_virtual(
|
||||
"distance_to_high_240d", "距一年高点", "价格位置",
|
||||
"收盘价 / 240日最高收盘价 - 1, 接近0代表贴近一年新高",
|
||||
deps=frozenset({"close"}), warmup_bars=241,
|
||||
),
|
||||
_virtual(
|
||||
"kdj_kd_diff", "KDJ K-D差", "趋势",
|
||||
"KDJ K值 - D值, 正值代表快线在慢线上方",
|
||||
deps=frozenset({"kdj_k", "kdj_d"}),
|
||||
),
|
||||
)
|
||||
|
||||
_REGISTRY: dict[str, FactorSpec] = {}
|
||||
|
||||
|
||||
def register_factor(spec: FactorSpec) -> None:
|
||||
"""注册因子; 重复 id 且版本未增时拒绝 (fail-closed)。"""
|
||||
existing = _REGISTRY.get(spec.id)
|
||||
if existing is not None and existing.version >= spec.version:
|
||||
raise ValueError(f"factor id 已注册且版本未提升: {spec.id}")
|
||||
_REGISTRY[spec.id] = spec
|
||||
|
||||
|
||||
for _spec in _CATALOG:
|
||||
register_factor(_spec)
|
||||
|
||||
|
||||
def get_factor(fid: str) -> FactorSpec | None:
|
||||
return _REGISTRY.get(fid)
|
||||
|
||||
|
||||
def unregister_factor(fid: str) -> FactorSpec | None:
|
||||
"""注销动态注册的因子 (内置目录因子不可注销, fail-closed)。"""
|
||||
if any(spec.id == fid for spec in _CATALOG):
|
||||
raise ValueError(f"内置因子不可注销: {fid}")
|
||||
return _REGISTRY.pop(fid, None)
|
||||
|
||||
|
||||
def _ordered_specs() -> list[FactorSpec]:
|
||||
"""内置目录顺序在前, 动态注册因子 (custom/composite) 按注册顺序追加。"""
|
||||
ordered: list[FactorSpec] = list(_CATALOG)
|
||||
known = {spec.id for spec in _CATALOG}
|
||||
ordered.extend(spec for fid, spec in _REGISTRY.items() if fid not in known)
|
||||
return ordered
|
||||
|
||||
|
||||
def all_factors(
|
||||
asset_type: str | None = None,
|
||||
stable_only: bool = False,
|
||||
) -> list[FactorSpec]:
|
||||
"""按目录顺序返回因子; asset_type 过滤适用资产, stable_only 过滤实验/废弃因子。"""
|
||||
return [
|
||||
spec for spec in _ordered_specs()
|
||||
if (asset_type is None or asset_type in spec.asset_types)
|
||||
and (not stable_only or spec.stability == "stable")
|
||||
]
|
||||
|
||||
|
||||
def factor_dependencies(fids) -> frozenset[str]:
|
||||
"""递归展开依赖到 enriched base 列; 未知 id 原样保留 (与 scoring_dependencies 历史语义一致)。"""
|
||||
resolved: set[str] = set()
|
||||
for fid in fids:
|
||||
spec = _REGISTRY.get(str(fid))
|
||||
if spec is None:
|
||||
resolved.add(str(fid))
|
||||
elif spec.dependencies:
|
||||
resolved.update(spec.dependencies)
|
||||
else:
|
||||
resolved.add(spec.id)
|
||||
return frozenset(resolved)
|
||||
|
||||
|
||||
def factor_columns_view() -> list[dict]:
|
||||
"""历史 FACTOR_COLUMNS 兼容视图 (顺序、键一致; 动态注册因子追加在末尾)。"""
|
||||
return [spec.column_view() for spec in _ordered_specs()]
|
||||
|
||||
|
||||
def virtual_dependencies() -> dict[str, frozenset[str]]:
|
||||
"""历史 VIRTUAL_SCORING_DEPENDENCIES 兼容视图。"""
|
||||
return {
|
||||
spec.id: spec.dependencies
|
||||
for spec in _CATALOG
|
||||
if spec.kind == "virtual" and spec.dependencies
|
||||
}
|
||||
|
||||
|
||||
def scoring_warmups() -> dict[str, int]:
|
||||
"""历史 _ROLLING_SCORING_WARMUP 兼容视图 (仅滚动窗口虚拟因子)。"""
|
||||
return {
|
||||
spec.id: spec.warmup_bars
|
||||
for spec in _CATALOG
|
||||
if spec.kind == "virtual" and spec.warmup_bars > 1
|
||||
}
|
||||
@@ -0,0 +1,190 @@
|
||||
"""自定义/复合因子存储 (P3) — data/user_data/custom_factors/*.json。
|
||||
|
||||
镜像 custom_signals 的持久化写法; 单文件损坏只禁用该因子并告警, 不影响启动
|
||||
(对齐 CONTRIBUTING 第 4 节插件隔离要求)。生命周期状态: draft → active →
|
||||
watch → retired (P4 状态机, 存储字段就绪, 迁移逻辑见巡检设计)。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import logging
|
||||
import re
|
||||
from datetime import datetime
|
||||
from pathlib import Path
|
||||
|
||||
from app.factors.dsl import compile_formula
|
||||
from app.factors.registry import FactorSpec, factor_dependencies, get_factor, register_factor
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
CUSTOM_ID_PATTERN = re.compile(r"^uf_[a-z0-9_]{1,40}$")
|
||||
COMPOSITE_ID_PATTERN = re.compile(r"^cf_[a-z0-9_]{1,40}$")
|
||||
MAX_COMPOSITE_MEMBERS = 8
|
||||
STATUSES = frozenset({"draft", "active", "watch", "retired"})
|
||||
|
||||
|
||||
def _dir(data_dir: Path) -> Path:
|
||||
directory = data_dir / "user_data" / "custom_factors"
|
||||
directory.mkdir(parents=True, exist_ok=True)
|
||||
return directory
|
||||
|
||||
|
||||
def _path(data_dir: Path, factor_id: str) -> Path:
|
||||
return _dir(data_dir) / f"{factor_id}.json"
|
||||
|
||||
|
||||
def load_all(data_dir: Path) -> list[dict]:
|
||||
"""读取全部自定义/复合因子定义; 损坏文件跳过。"""
|
||||
out: list[dict] = []
|
||||
for file in sorted(_dir(data_dir).glob("*.json")):
|
||||
try:
|
||||
out.append(json.loads(file.read_text(encoding="utf-8")))
|
||||
except Exception as exc:
|
||||
logger.warning("custom factor load failed %s: %s", file.name, exc)
|
||||
return out
|
||||
|
||||
|
||||
def save_one(data_dir: Path, definition: dict) -> None:
|
||||
target = _path(data_dir, str(definition["id"]))
|
||||
target.parent.mkdir(parents=True, exist_ok=True)
|
||||
target.write_text(json.dumps(definition, ensure_ascii=False, indent=2), encoding="utf-8")
|
||||
|
||||
|
||||
def delete_one(data_dir: Path, factor_id: str) -> bool:
|
||||
target = _path(data_dir, factor_id)
|
||||
if target.exists():
|
||||
target.unlink()
|
||||
return True
|
||||
return False
|
||||
|
||||
|
||||
def _now() -> str:
|
||||
return datetime.now().isoformat(timespec="seconds")
|
||||
|
||||
|
||||
def to_spec(definition: dict) -> FactorSpec:
|
||||
"""定义 → FactorSpec; 校验失败抛 ValueError (调用方 fail-closed)。
|
||||
|
||||
custom: 依赖/预热由 DSL 编译推导 (编译失败即拒绝注册)。
|
||||
composite: 依赖 = 成员递归展开; 预热 = 成员最大值; 循环引用拒绝。
|
||||
"""
|
||||
kind = str(definition.get("kind", "custom"))
|
||||
factor_id = str(definition.get("id", ""))
|
||||
label = str(definition.get("label", "")).strip()
|
||||
if not label:
|
||||
raise ValueError("label 不能为空")
|
||||
pattern = COMPOSITE_ID_PATTERN if kind == "composite" else CUSTOM_ID_PATTERN
|
||||
if not pattern.match(factor_id):
|
||||
raise ValueError(f"id 必须匹配 {pattern.pattern}")
|
||||
status = str(definition.get("status", "draft"))
|
||||
if status not in STATUSES:
|
||||
raise ValueError(f"status 必须是 {sorted(STATUSES)} 之一")
|
||||
|
||||
if kind == "custom":
|
||||
formula = str(definition.get("formula", ""))
|
||||
compiled = compile_formula(formula)
|
||||
if not compiled.ok:
|
||||
first = compiled.errors[0]
|
||||
raise ValueError(f"公式无效 [{first.code}]: {first.message}")
|
||||
return FactorSpec(
|
||||
id=factor_id,
|
||||
label=label,
|
||||
group=str(definition.get("group", "自定义")),
|
||||
formula_text=formula,
|
||||
kind="custom",
|
||||
version=int(definition.get("version", 1)),
|
||||
dependencies=frozenset(compiled.dependencies),
|
||||
warmup_bars=compiled.warmup_bars,
|
||||
direction=str(definition.get("direction", "none")), # type: ignore[arg-type]
|
||||
stability="stable" if status == "active" else "experimental",
|
||||
)
|
||||
|
||||
if kind != "composite":
|
||||
raise ValueError(f"未知 kind: {kind}")
|
||||
members_raw = definition.get("members")
|
||||
if not isinstance(members_raw, dict) or not (2 <= len(members_raw) <= MAX_COMPOSITE_MEMBERS):
|
||||
raise ValueError(f"composite 成员必须是 {2}~{MAX_COMPOSITE_MEMBERS} 个")
|
||||
from app.factors.dsl import BASE_COLUMNS
|
||||
|
||||
components: list[tuple[str, float]] = []
|
||||
for member_id, weight in members_raw.items():
|
||||
member_id = str(member_id)
|
||||
if member_id == factor_id:
|
||||
raise ValueError("composite 不能引用自身")
|
||||
try:
|
||||
weight = float(weight)
|
||||
except (TypeError, ValueError) as exc:
|
||||
raise ValueError(f"成员 {member_id} 权重必须是数字") from exc
|
||||
if not weight:
|
||||
raise ValueError(f"成员 {member_id} 权重不能为 0")
|
||||
# 成员 = 注册表因子 或 enriched 基准列 (已物化, 可直接参与组合)
|
||||
if get_factor(member_id) is None and member_id not in BASE_COLUMNS:
|
||||
raise ValueError(f"未知成员因子: {member_id}")
|
||||
components.append((member_id, weight))
|
||||
# 环检测沿 components 链走 (依赖已展开, 看不到链路成员)
|
||||
seen = {factor_id}
|
||||
frontier = [member_id for member_id, _ in components]
|
||||
while frontier:
|
||||
current = frontier.pop()
|
||||
if current in seen:
|
||||
raise ValueError("composite 成员存在循环引用")
|
||||
seen.add(current)
|
||||
current_spec = get_factor(current)
|
||||
if current_spec is not None and current_spec.kind == "composite":
|
||||
frontier.extend(member_id for member_id, _ in current_spec.components)
|
||||
dependencies = factor_dependencies([member_id for member_id, _ in components])
|
||||
warmup = max(
|
||||
((get_factor(member_id).warmup_bars if get_factor(member_id) else 1) for member_id, _ in components),
|
||||
default=1,
|
||||
)
|
||||
formula_text = " + ".join(
|
||||
f"{weight:g}*zscore({member_id})" for member_id, weight in components
|
||||
)
|
||||
return FactorSpec(
|
||||
id=factor_id,
|
||||
label=label,
|
||||
group=str(definition.get("group", "组合")),
|
||||
formula_text=formula_text,
|
||||
kind="composite",
|
||||
version=int(definition.get("version", 1)),
|
||||
dependencies=dependencies,
|
||||
warmup_bars=warmup,
|
||||
direction=str(definition.get("direction", "none")), # type: ignore[arg-type]
|
||||
components=tuple(components),
|
||||
stability="stable" if status == "active" else "experimental",
|
||||
)
|
||||
|
||||
|
||||
def register_definition(definition: dict) -> FactorSpec:
|
||||
"""定义 → spec → 注册 (重复 id 版本未升时由注册表拒绝)。"""
|
||||
spec = to_spec(definition)
|
||||
register_factor(spec)
|
||||
return spec
|
||||
|
||||
|
||||
def load_into_registry(data_dir: Path) -> list[str]:
|
||||
"""启动期把存储中的因子注册进注册表; 单个失败只跳过并告警。
|
||||
|
||||
多轮加载: composite 成员可能引用尚未加载的 custom/其他 composite (文件按
|
||||
字母序加载, cf_* 先于 uf_*), 失败的 composite 延后重试, 覆盖链式引用;
|
||||
重试用尽仍失败的只告警不阻塞启动。
|
||||
"""
|
||||
loaded: list[str] = []
|
||||
pending = list(load_all(data_dir))
|
||||
for round_index in range(3):
|
||||
deferred: list[dict] = []
|
||||
for definition in pending:
|
||||
try:
|
||||
register_definition(definition)
|
||||
loaded.append(str(definition["id"]))
|
||||
except ValueError as exc:
|
||||
if round_index < 2 and str(definition.get("kind")) == "composite":
|
||||
deferred.append(definition)
|
||||
else:
|
||||
logger.warning("custom factor 注册失败 %s: %s", definition.get("id"), exc)
|
||||
except Exception as exc:
|
||||
logger.warning("custom factor 注册失败 %s: %s", definition.get("id"), exc)
|
||||
if not deferred:
|
||||
break
|
||||
pending = deferred
|
||||
return loaded
|
||||
@@ -666,9 +666,12 @@ def compute_signals(df: pl.DataFrame, needed: set[str] | None = None) -> pl.Data
|
||||
if want:
|
||||
df = df.with_columns([expressions[name] for name in SIGNAL_DEPENDENCIES if name in want])
|
||||
|
||||
# 自定义信号（用户配置的字段+运算符+值组合，编译为布尔列）
|
||||
# 自定义信号（用户配置的字段+运算符+值组合，编译为布尔列）。
|
||||
# 条件引用的注册表因子列先复用评分物化管线补算 (虚拟/自定义/复合均可)。
|
||||
from app.strategy import custom_signals
|
||||
df = custom_signals.inject(df, _get_custom_signal_exprs(), needed=needed)
|
||||
exprs = _get_custom_signal_exprs()
|
||||
df = custom_signals.materialize_factor_columns(df, exprs, needed=needed)
|
||||
df = custom_signals.inject(df, exprs, needed=needed)
|
||||
|
||||
return df
|
||||
|
||||
@@ -1053,13 +1056,28 @@ def _select_storage_cols(df: pl.DataFrame) -> pl.DataFrame:
|
||||
|
||||
DEVIATION_WINDOWS: tuple[int, ...] = (3, 10, 30)
|
||||
|
||||
# 各交易所基准指数 (偏离值规则的「对应指数」近似): 优先分类指数, 缺失时回退
|
||||
# 各板块基准指数 (偏离值规则的「对应指数」, 按交易所官方口径): 优先首选, 缺失时回退
|
||||
# - 沪主板: 上证A指 → 上证指数 (两者差异可忽略)
|
||||
# - 科创板: 科创50 (上交所《交易规则》2026修订 6.12 指定基准) → 上证A指
|
||||
# - 深主板: 深证A指 → 深证成指 (深交所投教口径)
|
||||
# - 创业板: 创业板综合指数 → 深证A指 (深交所投教口径)
|
||||
# - 北交所: 北证50 → 上证指数 (北交所《交易规则》5.4.4)
|
||||
_BENCHMARK_PREFERENCE: dict[str, list[str]] = {
|
||||
"SH": ["000002.SH", "000001.SH"], # 上证A指 → 上证指数
|
||||
"SZ": ["399107.SZ", "399001.SZ"], # 深证A指 → 深证成指
|
||||
"BJ": ["899050.BJ", "000001.SH"], # 北证50 → 上证指数
|
||||
"SH": ["000002.SH", "000001.SH"],
|
||||
"STAR": ["000688.SH", "000002.SH"],
|
||||
"SZ": ["399107.SZ", "399001.SZ"],
|
||||
"GEM": ["399102.SZ", "399107.SZ"],
|
||||
"BJ": ["899050.BJ", "000001.SH"],
|
||||
}
|
||||
|
||||
# 偏离值计算需要的全部基准指数 (quote_service 并入实时显式拉取, 不依赖监控规则)
|
||||
BENCHMARK_INDEX_SYMBOLS: frozenset[str] = frozenset(
|
||||
sym for cands in _BENCHMARK_PREFERENCE.values() for sym in cands
|
||||
)
|
||||
|
||||
# 全部板块基准键 (SH/STAR/SZ/GEM/BJ)
|
||||
BENCH_KEYS: tuple[str, ...] = tuple(_BENCHMARK_PREFERENCE)
|
||||
|
||||
_benchmark_cache: dict[str, tuple[float, pl.DataFrame | None]] = {}
|
||||
_BENCHMARK_CACHE_TTL = 600.0
|
||||
|
||||
@@ -1067,7 +1085,8 @@ _BENCHMARK_CACHE_TTL = 600.0
|
||||
def load_benchmark_momentum(data_dir: Path) -> pl.DataFrame | None:
|
||||
"""读取指数日K, 计算各基准指数的滚动 N 日涨跌幅。
|
||||
|
||||
返回长表: date, bench_exchange, bench_close, bench_mom3d, bench_mom10d, bench_mom30d。
|
||||
返回长表: date, bench_key, bench_close, bench_mom3d, bench_mom10d, bench_mom30d。
|
||||
bench_key 为板块基准键 (SH/STAR/SZ/GEM/BJ, 见 _BENCHMARK_PREFERENCE)。
|
||||
bench_close 供盘中路径外推今日基准动量 (benchmark_momentum_today)。
|
||||
无可用指数数据时返回 None (偏离列置 null, 不阻塞主流程)。
|
||||
进程内按 data_dir 缓存 (TTL 10 分钟)。
|
||||
@@ -1085,11 +1104,11 @@ def load_benchmark_momentum(data_dir: Path) -> pl.DataFrame | None:
|
||||
index_glob = str(Path(data_dir) / "kline_index_daily" / "**" / "*.parquet")
|
||||
wanted: list[str] = []
|
||||
bench_of: dict[str, str] = {}
|
||||
for exchange, candidates in _BENCHMARK_PREFERENCE.items():
|
||||
for bench_key, candidates in _BENCHMARK_PREFERENCE.items():
|
||||
for sym in candidates:
|
||||
if sym not in bench_of:
|
||||
wanted.append(sym)
|
||||
bench_of[sym] = exchange
|
||||
bench_of[sym] = bench_key
|
||||
lf = scan_daily_parquet(
|
||||
index_glob, cast_options=pl.ScanCastOptions(integer_cast="allow-float")
|
||||
)
|
||||
@@ -1102,15 +1121,15 @@ def load_benchmark_momentum(data_dir: Path) -> pl.DataFrame | None:
|
||||
if not df_idx.is_empty():
|
||||
available = set(df_idx["symbol"].to_list())
|
||||
picked = [s for s in wanted if s in available]
|
||||
# 每个交易所取优先级最高的可用基准; 全缺时回退到任一可用基准。
|
||||
# 同一基准可服务多个交易所 (如北证50 缺失时北交所回退上证指数)。
|
||||
# 每个板块取优先级最高的可用基准; 全缺时回退到任一可用基准。
|
||||
# 同一基准可服务多个板块 (如科创50 缺失时科创板回退上证A指)。
|
||||
pairs: list[tuple[str, str]] = []
|
||||
for exchange, candidates in _BENCHMARK_PREFERENCE.items():
|
||||
for bench_key, candidates in _BENCHMARK_PREFERENCE.items():
|
||||
hit = next((s for s in candidates if s in available), None)
|
||||
if hit is None and picked:
|
||||
hit = picked[0]
|
||||
if hit is not None:
|
||||
pairs.append((hit, exchange))
|
||||
pairs.append((hit, bench_key))
|
||||
df_bench = df_idx.filter(pl.col("symbol").is_in([p[0] for p in pairs]))
|
||||
if not df_bench.is_empty():
|
||||
df_bench = df_bench.with_columns(
|
||||
@@ -1119,16 +1138,16 @@ def load_benchmark_momentum(data_dir: Path) -> pl.DataFrame | None:
|
||||
(pl.col("close") / pl.col("close").shift(n).over("symbol") - 1).alias(f"_bm{n}")
|
||||
for n in DEVIATION_WINDOWS
|
||||
]).rename({f"_bm{n}": f"bench_mom{n}d" for n in DEVIATION_WINDOWS})
|
||||
exchange_map = pl.DataFrame({
|
||||
key_map = pl.DataFrame({
|
||||
"symbol": [p[0] for p in pairs],
|
||||
"bench_exchange": [p[1] for p in pairs],
|
||||
"bench_key": [p[1] for p in pairs],
|
||||
})
|
||||
frame = (
|
||||
df_bench.join(exchange_map, on="symbol", how="inner")
|
||||
.select(["date", "bench_exchange", "close",
|
||||
df_bench.join(key_map, on="symbol", how="inner")
|
||||
.select(["date", "bench_key", "close",
|
||||
*[f"bench_mom{n}d" for n in DEVIATION_WINDOWS]])
|
||||
.rename({"close": "bench_close"})
|
||||
.unique(subset=["date", "bench_exchange"])
|
||||
.unique(subset=["date", "bench_key"])
|
||||
)
|
||||
except Exception as exc: # noqa: BLE001
|
||||
logger.warning("基准指数偏离数据加载失败: %s", exc)
|
||||
@@ -1138,14 +1157,21 @@ def load_benchmark_momentum(data_dir: Path) -> pl.DataFrame | None:
|
||||
return frame
|
||||
|
||||
|
||||
def _bench_exchange_expr() -> pl.Expr:
|
||||
"""symbol 后缀 → 交易所 (SH/SZ/BJ), 无法识别时 null。"""
|
||||
def _bench_key_expr() -> pl.Expr:
|
||||
"""symbol → 板块基准键 (SH/STAR/SZ/GEM/BJ), 无法识别时 null。
|
||||
|
||||
北交所按后缀; 沪市按 68 前缀区分科创板; 深市按 30 前缀区分创业板。
|
||||
与 abnormal_moves.board_of 的板块判定同口径。
|
||||
"""
|
||||
code = pl.col("symbol").str.slice(0, 6)
|
||||
suffix = pl.col("symbol").str.slice(-2).str.to_uppercase()
|
||||
return (
|
||||
pl.col("symbol").str.slice(-2).str.to_uppercase().replace(
|
||||
{ex: ex for ex in _BENCHMARK_PREFERENCE},
|
||||
default=None,
|
||||
return_dtype=pl.Utf8,
|
||||
)
|
||||
pl.when(suffix == "BJ").then(pl.lit("BJ"))
|
||||
.when((suffix == "SH") & code.str.starts_with("68")).then(pl.lit("STAR"))
|
||||
.when(suffix == "SH").then(pl.lit("SH"))
|
||||
.when((suffix == "SZ") & code.str.starts_with("30")).then(pl.lit("GEM"))
|
||||
.when(suffix == "SZ").then(pl.lit("SZ"))
|
||||
.otherwise(pl.lit(None, dtype=pl.Utf8))
|
||||
)
|
||||
|
||||
|
||||
@@ -1172,8 +1198,8 @@ def attach_deviation_columns(df: pl.DataFrame, data_dir: Path) -> pl.DataFrame:
|
||||
for n in missing
|
||||
])
|
||||
out = (
|
||||
df.with_columns(_bench_exchange_expr().alias("_bench_ex"))
|
||||
.join(bench, left_on=["_bench_ex", "date"], right_on=["bench_exchange", "date"], how="left")
|
||||
df.with_columns(_bench_key_expr().alias("_bench_ex"))
|
||||
.join(bench, left_on=["_bench_ex", "date"], right_on=["bench_key", "date"], how="left")
|
||||
.with_columns([
|
||||
(pl.col(f"momentum_{n}d") - pl.col(f"bench_mom{n}d")).alias(f"deviate_{n}d")
|
||||
for n in DEVIATION_WINDOWS
|
||||
@@ -1210,16 +1236,21 @@ def _bench_rt_pct_of(index_quotes: pl.DataFrame | None, candidates: list[str]) -
|
||||
return 0.0
|
||||
|
||||
|
||||
def bench_rt_pct_for(index_quotes: pl.DataFrame | None, bench_key: str) -> float:
|
||||
"""板块基准键的指数今日实时涨跌 (小数制), 供异动总览实时叠加等外部消费。"""
|
||||
return _bench_rt_pct_of(index_quotes, _BENCHMARK_PREFERENCE.get(bench_key, []))
|
||||
|
||||
|
||||
def benchmark_momentum_today(
|
||||
data_dir: Path,
|
||||
index_quotes: pl.DataFrame | None = None,
|
||||
) -> pl.DataFrame | None:
|
||||
"""各交易所基准指数的「今日」N 日动量 (盘中实时外推)。
|
||||
"""各板块基准指数的「今日」N 日动量 (盘中实时外推)。
|
||||
|
||||
基准日K parquet 盘中不含今日, 今日基准收盘 = 昨收 × (1 + 实时涨跌)。
|
||||
N 日动量 = 今日基准收盘 / N 个交易日前的收盘 - 1; 交易所与
|
||||
N 日动量 = 今日基准收盘 / N 个交易日前的收盘 - 1; 板块与
|
||||
load_benchmark_momentum 的选基逻辑一致 (同一 TTL 缓存帧)。
|
||||
返回小表: bench_exchange, bench_mom3d, bench_mom10d, bench_mom30d。
|
||||
返回小表: bench_key, bench_mom3d, bench_mom10d, bench_mom30d。
|
||||
无基准数据时 None。
|
||||
"""
|
||||
bench = load_benchmark_momentum(data_dir)
|
||||
@@ -1232,15 +1263,15 @@ def benchmark_momentum_today(
|
||||
if bench.is_empty():
|
||||
return None
|
||||
rows: list[dict[str, float | str]] = []
|
||||
for ex in sorted(bench["bench_exchange"].unique().to_list()):
|
||||
sub = bench.filter(pl.col("bench_exchange") == ex).sort("date")
|
||||
for k in sorted(bench["bench_key"].unique().to_list()):
|
||||
sub = bench.filter(pl.col("bench_key") == k).sort("date")
|
||||
closes = sub["bench_close"]
|
||||
if closes.len() == 0:
|
||||
continue
|
||||
yesterday_close = closes[-1]
|
||||
rt = _bench_rt_pct_of(index_quotes, _BENCHMARK_PREFERENCE.get(ex, []))
|
||||
rt = _bench_rt_pct_of(index_quotes, _BENCHMARK_PREFERENCE.get(k, []))
|
||||
row: dict[str, float | str] = {
|
||||
"bench_exchange": ex,
|
||||
"bench_key": k,
|
||||
}
|
||||
for n in DEVIATION_WINDOWS:
|
||||
base = closes[-n] if closes.len() >= n else None # N 个交易日前 (不含今日)
|
||||
@@ -1252,7 +1283,7 @@ def benchmark_momentum_today(
|
||||
rows.append(row)
|
||||
if not rows:
|
||||
return None
|
||||
schema = {"bench_exchange": pl.Utf8, **{f"bench_mom{n}d": pl.Float64 for n in DEVIATION_WINDOWS}}
|
||||
schema = {"bench_key": pl.Utf8, **{f"bench_mom{n}d": pl.Float64 for n in DEVIATION_WINDOWS}}
|
||||
return pl.DataFrame(rows, schema=schema)
|
||||
|
||||
|
||||
@@ -1283,8 +1314,8 @@ def attach_deviation_columns_today(
|
||||
for n in DEVIATION_WINDOWS
|
||||
]
|
||||
return (
|
||||
df.with_columns(_bench_exchange_expr().alias("_bench_ex"))
|
||||
.join(bench, left_on="_bench_ex", right_on="bench_exchange", how="left")
|
||||
df.with_columns(_bench_key_expr().alias("_bench_ex"))
|
||||
.join(bench, left_on="_bench_ex", right_on="bench_key", how="left")
|
||||
.with_columns(exprs)
|
||||
.drop(["_bench_ex", *[f"bench_mom{n}d" for n in DEVIATION_WINDOWS]])
|
||||
)
|
||||
|
||||
@@ -66,6 +66,48 @@ def _prune_partial_enriched_partitions(daily_dir: Path, enriched_dir: Path) -> l
|
||||
return pruned
|
||||
|
||||
|
||||
def _prune_stale_price_partitions(
|
||||
daily_dir: Path, enriched_dir: Path, max_dates: int = 5
|
||||
) -> list[str]:
|
||||
"""删除收盘价与官方日线不一致的 enriched 日期分区。
|
||||
|
||||
实时 flush 写入的当日分区行数与 daily 相同, 但收盘价可能停留在收盘集合
|
||||
竞价前的快照 (实测: TickFlow 实时端点收盘后仍长期返回旧价, 3392/5554 只
|
||||
股票当日收盘价与官方日线不符), #223 的行数校验识别不到。对最近若干交易日
|
||||
做值级比对: enriched.raw_close 与 daily.close 任一标的差超过半个最小报价
|
||||
单位即删分区, 由后续增量重算按官方日线全市场重建。
|
||||
"""
|
||||
import shutil
|
||||
|
||||
common = sorted(
|
||||
(
|
||||
p.stem.split("=", 1)[1]
|
||||
for p in enriched_dir.glob("date=*")
|
||||
if (daily_dir / p.stem).exists()
|
||||
),
|
||||
reverse=True,
|
||||
)[:max_dates]
|
||||
pruned: list[str] = []
|
||||
for ds in common:
|
||||
try:
|
||||
daily = pl.read_parquet(
|
||||
daily_dir / f"date={ds}" / "*.parquet", columns=["symbol", "close"]
|
||||
)
|
||||
enr = pl.read_parquet(
|
||||
enriched_dir / f"date={ds}" / "*.parquet", columns=["symbol", "raw_close"]
|
||||
)
|
||||
except Exception:
|
||||
continue # 列缺失/不可读 → 交给既有完整性检查兜底
|
||||
joined = enr.join(daily, on="symbol", how="inner").drop_nulls()
|
||||
if joined.is_empty():
|
||||
continue
|
||||
bad = joined.filter((pl.col("raw_close") - pl.col("close")).abs() > 0.005)
|
||||
if not bad.is_empty():
|
||||
shutil.rmtree(enriched_dir / f"date={ds}", ignore_errors=True)
|
||||
pruned.append(ds)
|
||||
return pruned
|
||||
|
||||
|
||||
class PipelineStageError(RuntimeError):
|
||||
"""管道有阶段软失败(数据可能陈旧)时抛出, 让上层 job_store 把任务标记为 failed。
|
||||
|
||||
@@ -388,14 +430,19 @@ def run_now(
|
||||
daily_days = len(list(daily_dir.glob("date=*"))) if daily_dir.exists() else 0
|
||||
prev_enriched_days = len(list(enriched_dir.glob("date=*"))) if enriched_exists else 0
|
||||
|
||||
# 部分分区修复 (#223): 删除被实时合并提前创建、覆盖不全的 enriched 分区,
|
||||
# 让下方计数比较与增量计算把它们重新当新日期处理
|
||||
# 部分分区修复 (#223) + 收盘价过期分区修复: 删除被实时合并提前创建、覆盖不全
|
||||
# 或收盘价停留在竞价前快照的 enriched 分区, 让下方计数比较与增量计算把它们
|
||||
# 重新当新日期处理 (值级比对以官方日线为准, 实时源不纠错也能自愈)
|
||||
if enriched_exists:
|
||||
partial_pruned = _prune_partial_enriched_partitions(daily_dir, enriched_dir)
|
||||
if partial_pruned:
|
||||
stale_pruned = _prune_stale_price_partitions(daily_dir, enriched_dir)
|
||||
pruned_dates = sorted(set(partial_pruned) | set(stale_pruned))
|
||||
if pruned_dates:
|
||||
logger.warning(
|
||||
"compute_enriched: 发现 %d 个覆盖不全的 enriched 分区, 已删除待重算: %s",
|
||||
len(partial_pruned), ", ".join(sorted(partial_pruned)[:10]),
|
||||
"compute_enriched: 发现 %d 个异常 enriched 分区 (覆盖不全 %d / 收盘价过期 %d), "
|
||||
"已删除待重算: %s",
|
||||
len(pruned_dates), len(partial_pruned), len(stale_pruned),
|
||||
", ".join(pruned_dates[:10]),
|
||||
)
|
||||
enriched_exists = enriched_dir.exists() and any(enriched_dir.glob("date=*"))
|
||||
prev_enriched_days = len(list(enriched_dir.glob("date=*"))) if enriched_exists else 0
|
||||
|
||||
@@ -19,6 +19,7 @@ from app.api import (
|
||||
backtest,
|
||||
data,
|
||||
ext_data,
|
||||
factors,
|
||||
financials,
|
||||
indices,
|
||||
intraday,
|
||||
@@ -104,6 +105,15 @@ async def _application_lifespan(app: FastAPI):
|
||||
repo = KlineRepository(store)
|
||||
app.state.datastore = store
|
||||
app.state.repo = repo
|
||||
# 自定义/复合因子载入注册表 (P3); 单个失败只跳过该因子 (fail-隔离)
|
||||
from app.factors.store import load_into_registry
|
||||
|
||||
try:
|
||||
loaded_factors = load_into_registry(store.data_dir)
|
||||
if loaded_factors:
|
||||
logger.info("custom factors loaded: %s", len(loaded_factors))
|
||||
except Exception as exc: # noqa: BLE001
|
||||
logger.warning("custom factors load failed: %s", exc)
|
||||
from app.services.mining_manager import MiningJobManager
|
||||
|
||||
mining_manager = MiningJobManager(store.data_dir)
|
||||
@@ -455,6 +465,7 @@ app.include_router(kline.router)
|
||||
app.include_router(watchlist.router)
|
||||
app.include_router(screener.router)
|
||||
app.include_router(backtest.router)
|
||||
app.include_router(factors.router)
|
||||
app.include_router(mining.router)
|
||||
app.include_router(intraday.router)
|
||||
app.include_router(indices.router)
|
||||
|
||||
@@ -27,7 +27,7 @@ from typing import Any
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.indicators.pipeline import DEVIATION_WINDOWS
|
||||
from app.indicators.pipeline import BENCH_KEYS, DEVIATION_WINDOWS, bench_rt_pct_for
|
||||
|
||||
# ── 规则表 ────────────────────────────────────────────────
|
||||
|
||||
@@ -57,6 +57,18 @@ RULES_META: list[dict[str, Any]] = [
|
||||
_BENCH_RT_CANDIDATES = ["000002.SH", "000001.SH", "399107.SZ", "399001.SZ", "899050.BJ"]
|
||||
|
||||
|
||||
def _bench_key_of(symbol: str) -> str:
|
||||
"""symbol → 板块基准键, 与 pipeline._bench_key_expr 同口径 (SH/STAR/SZ/GEM/BJ)。"""
|
||||
code = symbol.split(".")[0]
|
||||
if symbol.endswith(".BJ"):
|
||||
return "BJ"
|
||||
if symbol.endswith(".SH"):
|
||||
return "STAR" if code.startswith("68") else "SH"
|
||||
if symbol.endswith(".SZ"):
|
||||
return "GEM" if code.startswith("30") else "SZ"
|
||||
return ""
|
||||
|
||||
|
||||
def board_of(symbol: str) -> str:
|
||||
"""按代码前缀判定板块。"""
|
||||
code = symbol.split(".")[0]
|
||||
@@ -174,6 +186,15 @@ def build_overview(
|
||||
hist_rows: dict[str, dict[str, Any]] = hist["rows"]
|
||||
|
||||
bench_rt = _bench_rt_pct(quote_service) if quote_service is not None else 0.0
|
||||
# 实时叠加按板块基准: 科创板减科创50、创业板减创业板综指, 不再全市场混均值
|
||||
bench_by_key: dict[str, float] = {}
|
||||
if quote_service is not None:
|
||||
try:
|
||||
index_quotes = quote_service.get_index_quotes()
|
||||
except Exception:
|
||||
index_quotes = None
|
||||
for k in BENCH_KEYS:
|
||||
bench_by_key[k] = bench_rt_pct_for(index_quotes, k)
|
||||
# enriched 已含今日收盘 (盘后已同步) 时, 今日涨跌已计入历史偏离, 不再叠加
|
||||
includes_today = cache_date is not None and cache_date >= date.today().isoformat()
|
||||
|
||||
@@ -181,7 +202,9 @@ def build_overview(
|
||||
for symbol, base in hist_rows.items():
|
||||
rule = rule_for(symbol, base.get("name"))
|
||||
rt_pct = base.get("rt_pct")
|
||||
rt_delta = 0.0 if includes_today else ((rt_pct or 0.0) - bench_rt)
|
||||
rt_delta = 0.0 if includes_today else (
|
||||
(rt_pct or 0.0) - bench_by_key.get(_bench_key_of(symbol), 0.0)
|
||||
)
|
||||
|
||||
windows: dict[str, dict[str, Any]] = {}
|
||||
max_closeness = 0.0
|
||||
|
||||
@@ -0,0 +1,193 @@
|
||||
"""自动挖掘 L1 编排: 全量因子统计筛选 → 达标池。
|
||||
|
||||
流程定位 (对应方案「四层漏斗」):
|
||||
- L1 本模块: 注册表全量因子批量检验, 按置信档门槛筛出达标因子 (近期窗口,
|
||||
仅作"有信号"的先验过滤; 最终达标由挖掘引擎的逐折训练选择与嵌套样本外
|
||||
验证裁定)。
|
||||
- L2/L3/L4 由现有挖掘引擎完成: 相关性剪枝 (prune_correlated_factors)、
|
||||
束搜索组合 (beam_search_factor_combinations)、嵌套样本外验证与达标
|
||||
门槛 (evaluate_candidate_gate), 本模块不重复实现。
|
||||
|
||||
达标判据与检验页服务端判读同源 (|t_NW| / BH q / |IC| / |IR|), 按档放宽或收紧;
|
||||
q 值缺失时按"通过"处理 (探索档小样本下 BH 校正保守)。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import math
|
||||
import time
|
||||
from dataclasses import dataclass
|
||||
from datetime import date, timedelta
|
||||
from typing import Any, Literal
|
||||
|
||||
from app.backtest.factor import FactorBacktestService, FactorBatchConfig
|
||||
from app.factors.registry import factor_columns_view
|
||||
|
||||
Profile = Literal["exploratory", "balanced", "strict"]
|
||||
|
||||
# 挖掘请求的因子池上限 (与 MiningStartRequest.factor_names max_length 对齐)
|
||||
MAX_AUTO_POOL = 48
|
||||
|
||||
# L1 筛选窗口: 近一年 (与挖掘窗口解耦, 只筛"近期有信号", 长窗口验证交给引擎)
|
||||
SCREEN_WINDOW_DAYS = 365
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class ScreenGate:
|
||||
min_abs_ic: float
|
||||
min_abs_ir: float
|
||||
min_abs_t: float
|
||||
max_q: float
|
||||
|
||||
def to_dict(self) -> dict[str, float]:
|
||||
return {
|
||||
"min_abs_ic": self.min_abs_ic,
|
||||
"min_abs_ir": self.min_abs_ir,
|
||||
"min_abs_t": self.min_abs_t,
|
||||
"max_q": self.max_q,
|
||||
}
|
||||
|
||||
|
||||
SCREEN_GATES: dict[str, ScreenGate] = {
|
||||
"exploratory": ScreenGate(min_abs_ic=0.02, min_abs_ir=0.15, min_abs_t=1.5, max_q=0.20),
|
||||
"balanced": ScreenGate(min_abs_ic=0.02, min_abs_ir=0.30, min_abs_t=2.0, max_q=0.10),
|
||||
"strict": ScreenGate(min_abs_ic=0.03, min_abs_ir=0.50, min_abs_t=2.5, max_q=0.05),
|
||||
}
|
||||
|
||||
|
||||
def classify_factor(item: dict[str, Any], gate: ScreenGate) -> str | None:
|
||||
"""返回 None 表示达标; 否则返回首个未过的门槛, 格式统一为「类别 (细节)」。"""
|
||||
if item.get("error"):
|
||||
return f"计算失败 ({str(item['error'])[:40]})"
|
||||
ic = item.get("ic_mean")
|
||||
ir = item.get("ir")
|
||||
t = item.get("t_newey_west")
|
||||
q = item.get("q_value")
|
||||
if ic is None or ir is None:
|
||||
return "样本不足 (无有效 IC/IR)"
|
||||
if abs(ic) < gate.min_abs_ic:
|
||||
return f"预测力弱 (|IC|<{gate.min_abs_ic:.2f})"
|
||||
if abs(ir) < gate.min_abs_ir:
|
||||
return f"稳定度低 (|IR|<{gate.min_abs_ir:.2f})"
|
||||
if t is None:
|
||||
return "样本不足 (无 NW t 值)"
|
||||
if abs(t) < gate.min_abs_t:
|
||||
return f"不显著 (|t|<{gate.min_abs_t:.1f})"
|
||||
if q is not None and q > gate.max_q:
|
||||
return f"多重检验未过 (q>{gate.max_q:.2f})"
|
||||
return None
|
||||
|
||||
|
||||
def _short_reason(reason: str) -> str:
|
||||
"""失败原因归并到短类别 (「类别 (细节)」的前半段), 供原因分布统计。"""
|
||||
return reason.split(" (", 1)[0].strip()
|
||||
|
||||
|
||||
def _finite_or_none(value: Any) -> float | None:
|
||||
"""NaN/Inf 一律归 None, 避免写入任务存储时产生非法 JSON。"""
|
||||
if isinstance(value, (int, float)) and math.isfinite(value):
|
||||
return float(value)
|
||||
return None
|
||||
|
||||
|
||||
def _metric_row(item: dict[str, Any]) -> dict[str, Any]:
|
||||
return {
|
||||
"factor_name": item.get("factor_name"),
|
||||
"label": item.get("label") or item.get("factor_name"),
|
||||
"group": item.get("group") or "",
|
||||
"ic": _finite_or_none(item.get("ic_mean")),
|
||||
"ir": _finite_or_none(item.get("ir")),
|
||||
"t": _finite_or_none(item.get("t_newey_west")),
|
||||
"q": _finite_or_none(item.get("q_value")),
|
||||
"direction": 1 if (item.get("ic_mean") or 0) >= 0 else -1,
|
||||
}
|
||||
|
||||
|
||||
def screen_all_factors(
|
||||
engine: Any,
|
||||
*,
|
||||
asset_type: str,
|
||||
start: date | None,
|
||||
end: date,
|
||||
profile: str,
|
||||
max_factors: int = MAX_AUTO_POOL,
|
||||
) -> dict[str, Any]:
|
||||
"""L1 全量筛选: 注册表全部适用因子批量检验 → 达标池 + 失败原因分布。
|
||||
|
||||
start=None 时取近 SCREEN_WINDOW_DAYS 天; 显式 start 只会收紧 (不放宽) 筛选窗口。
|
||||
"""
|
||||
gate = SCREEN_GATES.get(profile)
|
||||
if gate is None:
|
||||
raise ValueError(f"unknown mining profile: {profile}")
|
||||
|
||||
candidates = [
|
||||
str(item["id"])
|
||||
for item in factor_columns_view()
|
||||
if asset_type in item.get("asset_types", ["stock"])
|
||||
]
|
||||
screen_start = max(start or date.min, end - timedelta(days=SCREEN_WINDOW_DAYS))
|
||||
began = time.perf_counter()
|
||||
service = FactorBacktestService(engine)
|
||||
batch = service.run_batch(FactorBatchConfig(
|
||||
factor_names=candidates,
|
||||
symbols=None,
|
||||
start=screen_start,
|
||||
end=end,
|
||||
rebalance="daily",
|
||||
asset_type=asset_type,
|
||||
))
|
||||
elapsed_ms = round((time.perf_counter() - began) * 1000, 1)
|
||||
|
||||
qualified: list[dict[str, Any]] = []
|
||||
failed: list[dict[str, Any]] = []
|
||||
by_name = {str(getattr(item, "factor_name", None)): item for item in batch.results}
|
||||
for name in candidates:
|
||||
item = by_name.get(name)
|
||||
if item is None:
|
||||
failed.append({"factor_name": name, "label": name, "group": "",
|
||||
"ic": None, "ir": None, "t": None, "q": None,
|
||||
"reason": "未返回结果"})
|
||||
continue
|
||||
# 非有限值先清洗 (NaN 与任何比较均为 False, 会绕过门槛误判达标)
|
||||
reason = classify_factor({
|
||||
"error": getattr(item, "error", None),
|
||||
"ic_mean": _finite_or_none(getattr(item, "ic_mean", None)),
|
||||
"ir": _finite_or_none(getattr(item, "ir", None)),
|
||||
"t_newey_west": _finite_or_none(getattr(item, "t_newey_west", None)),
|
||||
"q_value": _finite_or_none(getattr(item, "q_value", None)),
|
||||
}, gate)
|
||||
row = _metric_row({
|
||||
"factor_name": getattr(item, "factor_name", None),
|
||||
"label": getattr(item, "label", None),
|
||||
"group": getattr(item, "group", None),
|
||||
"ic_mean": getattr(item, "ic_mean", None),
|
||||
"ir": getattr(item, "ir", None),
|
||||
"t_newey_west": getattr(item, "t_newey_west", None),
|
||||
"q_value": getattr(item, "q_value", None),
|
||||
})
|
||||
if reason is None:
|
||||
qualified.append(row)
|
||||
else:
|
||||
failed.append({**row, "reason": reason})
|
||||
|
||||
# 池按 |IC|*|IR| 降序 (截面信噪比口径), 截断到挖掘上限
|
||||
qualified.sort(key=lambda row: abs(row["ic"] or 0.0) * abs(row["ir"] or 0.0), reverse=True)
|
||||
pool = [row["factor_name"] for row in qualified[:max_factors]]
|
||||
|
||||
reason_counts: dict[str, int] = {}
|
||||
for row in failed:
|
||||
category = _short_reason(row["reason"])
|
||||
reason_counts[category] = reason_counts.get(category, 0) + 1
|
||||
|
||||
return {
|
||||
"profile": profile,
|
||||
"gate": gate.to_dict(),
|
||||
"screen_window": {"start": screen_start.isoformat(), "end": end.isoformat()},
|
||||
"n_total": len(candidates),
|
||||
"n_qualified": len(qualified),
|
||||
"pool": pool,
|
||||
"pool_truncated": len(qualified) > len(pool),
|
||||
"qualified": qualified,
|
||||
"failed": failed,
|
||||
"reason_counts": dict(sorted(reason_counts.items(), key=lambda kv: -kv[1])),
|
||||
"elapsed_ms": elapsed_ms,
|
||||
}
|
||||
@@ -33,7 +33,7 @@ from datetime import date, datetime, time as dt_time
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.market_time import cn_now, cn_today
|
||||
from app.market_time import CN_TZ, cn_now, cn_today
|
||||
from app.parquet import scan_daily_parquet
|
||||
from app.services.index_const import CORE_INDEX_SYMBOLS
|
||||
from app.strategy.intraday_signals import IntradaySignalEvaluator
|
||||
@@ -46,6 +46,15 @@ SOURCE_LABELS = {
|
||||
"volume_delta": "放量", "abnormal": "异动", "date": "日期提醒",
|
||||
}
|
||||
|
||||
# final 定版确认容差: 快照时间戳允许早于边界 5s 内 (供应商时间戳精度不一)
|
||||
_FINAL_CONFIRM_SLACK_MS = 5_000
|
||||
|
||||
# final 定版边界与重试窗口终点 (北京时间)。收盘窗口终点 15:30, 恰与盘后管道
|
||||
# 启动同时: 管道运行期间轮询本就被暂停, 此后未确认的定版不再写盘, 当日分区
|
||||
# 由管道按官方日线值级校正 —— 避免定版重试与权威重建互相覆盖。
|
||||
_FINAL_BOUNDARY = {"morning_final": dt_time(11, 30), "close_final": dt_time(15, 0)}
|
||||
_FINAL_DEADLINE = {"morning_final": dt_time(12, 10), "close_final": dt_time(15, 30)}
|
||||
|
||||
|
||||
def _body_with_quote(body: str, ev: dict) -> str:
|
||||
"""推送正文尾部补上触发时的现价/涨跌幅 (日期提醒无行情, 自然为空)。
|
||||
@@ -240,6 +249,8 @@ class QuoteService:
|
||||
# 午休/收盘最终同步状态: 到边界后必须成功拉取一版行情, 再进入休盘态。
|
||||
self._final_sync_done: set[tuple[date, str]] = set()
|
||||
self._final_sync_failed: dict[tuple[date, str], str] = {}
|
||||
# 最近一次 final 定版拉取是否取得边界后快照 (None=非 final 拉取)
|
||||
self._last_final_confirmed: bool | None = None
|
||||
self._holiday_active = False # 交易日探针当前是否判休市 (日志去重)
|
||||
# 轮询放量 (volume_delta 规则): 上一轮全市场股票快照的 (累计成交量[手], 累计成交额[元])。
|
||||
# 每轮全量快照后更新 (含非连续竞价时段, 保证 13:00 恢复时 prev 是 12:59
|
||||
@@ -562,8 +573,17 @@ class QuoteService:
|
||||
}
|
||||
|
||||
def refresh(self) -> dict:
|
||||
"""手动触发一次行情拉取。"""
|
||||
self._fetch_quotes()
|
||||
"""手动触发一次行情拉取。
|
||||
|
||||
午休/收盘定版阶段同样走边界确认: 避免盘后手动刷新把竞价前的陈旧收盘价
|
||||
重新写回当日分区, 覆盖盘后管道按官方日线重建的结果。
|
||||
"""
|
||||
phase = self._market_phase()
|
||||
is_final = phase in {"morning_final", "close_final"}
|
||||
self._fetch_quotes(
|
||||
final=is_final,
|
||||
final_boundary_ms=self._final_boundary_ms(phase) if is_final else None,
|
||||
)
|
||||
return self.status()
|
||||
|
||||
# ================================================================
|
||||
@@ -579,16 +599,36 @@ class QuoteService:
|
||||
phase = self._market_phase()
|
||||
if self._should_fetch_for_phase(phase):
|
||||
is_final = phase in {"morning_final", "close_final"}
|
||||
ok = self._fetch_quotes(final=is_final)
|
||||
ok = self._fetch_quotes(
|
||||
final=is_final,
|
||||
final_boundary_ms=self._final_boundary_ms(phase),
|
||||
)
|
||||
if is_final:
|
||||
key = self._final_sync_key(phase)
|
||||
if key and ok:
|
||||
label = "午休" if phase == "morning_final" else "收盘"
|
||||
if key and ok and self._last_final_confirmed:
|
||||
self._final_sync_done.add(key)
|
||||
self._final_sync_failed.pop(key, None)
|
||||
logger.info("%s 最终行情同步完成, 进入休盘态", "午休" if phase == "morning_final" else "收盘")
|
||||
logger.info("%s 最终行情同步完成 (快照时间戳已达边界), 进入休盘态", label)
|
||||
elif key and self._past_final_deadline(phase):
|
||||
# 重试窗口结束仍未取得边界后快照: 接受现状停止轮询。
|
||||
# 实测有实时源收盘后长期返回竞价前旧价 (快照时间戳可信但价格不更新),
|
||||
# 此时盲目落盘只会固化旧价 —— 交由 15:30 盘后管道按官方日线校正。
|
||||
self._final_sync_done.add(key)
|
||||
self._final_sync_failed[key] = (
|
||||
"fetch_failed" if not ok else "unconfirmed_snapshot"
|
||||
)
|
||||
logger.warning(
|
||||
"%s 定版窗口结束仍未取得边界后快照 (%s), 停止轮询; "
|
||||
"当日分区由盘后管道按官方日线值级校正",
|
||||
label, "拉取失败" if not ok else "快照未确认",
|
||||
)
|
||||
elif key:
|
||||
self._final_sync_failed[key] = "fetch_failed"
|
||||
logger.warning("%s 最终行情同步失败, 将继续重试", "午休" if phase == "morning_final" else "收盘")
|
||||
self._final_sync_failed[key] = (
|
||||
"fetch_failed" if not ok else "unconfirmed_snapshot"
|
||||
)
|
||||
if not ok:
|
||||
logger.warning("%s 最终行情同步失败, 将继续重试", label)
|
||||
else:
|
||||
logger.debug("非轮询阶段(%s), 跳过行情轮询", phase)
|
||||
except Exception as e: # noqa: BLE001
|
||||
@@ -599,16 +639,20 @@ class QuoteService:
|
||||
time.sleep(0.5)
|
||||
waited += 0.5
|
||||
|
||||
def _fetch_quotes(self, *, final: bool = False) -> bool:
|
||||
"""拉取行情。加锁串行化 (后台轮询 vs 手动 refresh)。返回本轮是否成功更新。"""
|
||||
def _fetch_quotes(self, *, final: bool = False, final_boundary_ms: int | None = None) -> bool:
|
||||
"""拉取行情。加锁串行化 (后台轮询 vs 手动 refresh)。返回本轮是否成功更新。
|
||||
|
||||
final_boundary_ms: final 定版的边界时间戳 (ms)。传入时快照时间戳未达边界
|
||||
的本轮不落盘 (见 _process_full_market_records)。
|
||||
"""
|
||||
with self._fetch_lock:
|
||||
before = self._fetched_at
|
||||
if final:
|
||||
logger.info("最终行情同步开始")
|
||||
self._fetch_full_market_quotes()
|
||||
self._fetch_full_market_quotes(final_boundary_ms=final_boundary_ms)
|
||||
return self._fetched_at > before
|
||||
|
||||
def _fetch_full_market_quotes(self) -> None:
|
||||
def _fetch_full_market_quotes(self, final_boundary_ms: int | None = None) -> None:
|
||||
"""拉取全市场行情 → 写 daily + 计算 enriched + 更新缓存。"""
|
||||
from app.services import preferences
|
||||
|
||||
@@ -627,7 +671,15 @@ class QuoteService:
|
||||
replace_index_cache = True
|
||||
fetch_indices = getattr(provider, "get_realtime_indices", None)
|
||||
if callable(fetch_indices):
|
||||
wanted = sorted(set(CORE_INDEX_SYMBOLS) | self._collect_monitor_index_symbols())
|
||||
# 偏离值基准指数 (科创50/创业板综指等) 一并拉取, 供盘中
|
||||
# attach_deviation_columns_today 实时外推; 展示层仍按核心
|
||||
# 四只过滤, 多拉的指数不进侧栏。
|
||||
from app.indicators.pipeline import BENCHMARK_INDEX_SYMBOLS
|
||||
wanted = sorted(
|
||||
set(CORE_INDEX_SYMBOLS)
|
||||
| BENCHMARK_INDEX_SYMBOLS
|
||||
| self._collect_monitor_index_symbols()
|
||||
)
|
||||
try:
|
||||
fetched_indices = fetch_indices(wanted)
|
||||
if fetched_indices is None:
|
||||
@@ -645,6 +697,7 @@ class QuoteService:
|
||||
t0=t0,
|
||||
now_ts=now_ts,
|
||||
replace_index_cache=replace_index_cache,
|
||||
final_boundary_ms=final_boundary_ms,
|
||||
)
|
||||
return
|
||||
# 自定义源未配置 realtime → 回退 TickFlow
|
||||
@@ -684,8 +737,11 @@ class QuoteService:
|
||||
logger.info("拉取全市场行情 (universes=%s, SDK超时=30s×重试3)", universes)
|
||||
resp.extend(tf.quotes.get_by_universes(universes=universes) or [])
|
||||
logger.info("全市场行情拉取完成: %d 条 (%.2fs)", len(resp), time.perf_counter() - _u0)
|
||||
# 指数: 固定核心四只 + 监控规则标的, 按码显式拉取
|
||||
_core_syms = sorted(core_index_symbols | monitor_index_symbols)
|
||||
# 指数: 固定核心四只 + 偏离值基准指数 + 监控规则标的, 按码显式拉取
|
||||
from app.indicators.pipeline import BENCHMARK_INDEX_SYMBOLS
|
||||
_core_syms = sorted(
|
||||
core_index_symbols | BENCHMARK_INDEX_SYMBOLS | monitor_index_symbols
|
||||
)
|
||||
if _core_syms:
|
||||
_i0 = time.perf_counter()
|
||||
resp.extend(tf.quotes.get(symbols=_core_syms) or [])
|
||||
@@ -730,7 +786,9 @@ class QuoteService:
|
||||
"session": q.get("session"),
|
||||
})
|
||||
|
||||
self._process_full_market_records(records, t0=t0, now_ts=now_ts)
|
||||
self._process_full_market_records(
|
||||
records, t0=t0, now_ts=now_ts, final_boundary_ms=final_boundary_ms
|
||||
)
|
||||
|
||||
def _process_full_market_records(
|
||||
self,
|
||||
@@ -739,8 +797,14 @@ class QuoteService:
|
||||
t0: float,
|
||||
now_ts: float,
|
||||
replace_index_cache: bool = True,
|
||||
final_boundary_ms: int | None = None,
|
||||
) -> None:
|
||||
"""把全市场 records 写盘并增量计算 enriched。"""
|
||||
"""把全市场 records 写盘并增量计算 enriched。
|
||||
|
||||
final_boundary_ms (final 定版边界) 传入时, 快照最大时间戳未达边界的本轮
|
||||
只更新展示缓存, 不写 daily/enriched、不评估监控 —— 防止收盘后数据源仍
|
||||
返回竞价前旧价时把陈旧收盘价固化到当日分区。
|
||||
"""
|
||||
from app.services import preferences
|
||||
all_index_symbols = set(self._repo.get_index_symbol_set()) if self._repo else set()
|
||||
core_index_symbols = set(CORE_INDEX_SYMBOLS)
|
||||
@@ -755,6 +819,16 @@ class QuoteService:
|
||||
logger.warning("行情数据为空")
|
||||
return
|
||||
|
||||
# ---- final 定版确认: 快照最大时间戳达到边界 (含容差) 才允许落盘 ----
|
||||
confirmed_final: bool | None = None
|
||||
if final_boundary_ms is not None:
|
||||
ts_vals = [t for t in (r.get("timestamp") for r in records) if t]
|
||||
max_ts = max(ts_vals) if ts_vals else None
|
||||
confirmed_final = bool(
|
||||
max_ts is not None and max_ts >= final_boundary_ms - _FINAL_CONFIRM_SLACK_MS
|
||||
)
|
||||
self._last_final_confirmed = confirmed_final
|
||||
|
||||
index_records = [r for r in records if r.get("symbol") in all_index_symbols]
|
||||
etf_records = [r for r in records if r.get("symbol") in all_etf_symbols]
|
||||
stock_records = [
|
||||
@@ -781,6 +855,16 @@ class QuoteService:
|
||||
_persist_last_fetch(fetched_at)
|
||||
logger.info("行情刷新: %d 只股票, %d 只ETF, %d 只指数, 耗时 %.0fms", len(stock_records), len(etf_records), len(index_records), fetch_ms)
|
||||
|
||||
if confirmed_final is False:
|
||||
# 边界前的陈旧快照: 展示缓存已更新, 落盘与监控评估留待边界后快照。
|
||||
# 轮询线程会在定版窗口内持续重试, 窗口结束由 _poll_loop 放弃并告警。
|
||||
logger.info(
|
||||
"final 快照未达定版边界 (max quote_ts=%s, 边界=%s), 本轮跳过落盘",
|
||||
max_ts, final_boundary_ms,
|
||||
)
|
||||
self._broadcast_quote_updated()
|
||||
return
|
||||
|
||||
# 轮询放量状态更新 (volume_delta 规则的差值来源)
|
||||
self._update_volume_delta(stock_records, fetched_at)
|
||||
|
||||
@@ -868,6 +952,18 @@ class QuoteService:
|
||||
result = df.select(select_exprs).with_columns(
|
||||
pl.lit(cn_today()).cast(pl.Date).alias("date"),
|
||||
)
|
||||
# 停牌股回归: 实时源对停牌标的返回停牌前最后一份快照 — OHLCV 全为旧日
|
||||
# 真实值, 仅 timestamp 停在旧日。这类记录不属于当日, 不过滤会把旧日 K 线
|
||||
# 原样复制成当日假蜡烛 (如 301266.SZ 2026-09-04)。按 quote_ts 的北京
|
||||
# 日期归属过滤; 时间戳缺失/为空的源无法判断, 维持原行为保留。
|
||||
if "quote_ts" in result.columns:
|
||||
day_start_ms = int(
|
||||
datetime.combine(cn_today(), dt_time(0, 0), tzinfo=CN_TZ).timestamp() * 1000
|
||||
)
|
||||
result = result.filter(
|
||||
pl.col("quote_ts").is_null()
|
||||
| pl.col("quote_ts").is_between(day_start_ms, day_start_ms + 86_400_000, closed="left")
|
||||
)
|
||||
# 停牌/尚无集合竞价的记录 open/high 均为 0。必须在下方用 close 填充前
|
||||
# 过滤, 否则零成交行会被伪装成有效日K, 并在 batch 同步后作为实时残留
|
||||
# 反复触发历史完整性修复。
|
||||
@@ -979,6 +1075,20 @@ class QuoteService:
|
||||
return (cn_today(), "close")
|
||||
return None
|
||||
|
||||
@classmethod
|
||||
def _final_boundary_ms(cls, phase: str) -> int | None:
|
||||
"""final 阶段定版边界的 epoch ms (按北京时间当日换算, 不依赖服务器时区)。"""
|
||||
b = _FINAL_BOUNDARY.get(phase)
|
||||
if b is None:
|
||||
return None
|
||||
return int(datetime.combine(cn_today(), b, tzinfo=CN_TZ).timestamp() * 1000)
|
||||
|
||||
@classmethod
|
||||
def _past_final_deadline(cls, phase: str) -> bool:
|
||||
"""是否已过 final 重试窗口终点 (用于放弃未确认的定版重试)。"""
|
||||
dl = _FINAL_DEADLINE.get(phase)
|
||||
return dl is not None and cn_now().time() >= dl
|
||||
|
||||
def _holiday_gate(self) -> bool:
|
||||
"""交易日探针门控: 确定休市 → False (停止轮询, 含 final 定版)。
|
||||
|
||||
|
||||
@@ -66,6 +66,49 @@ _OP_BUILDERS = {
|
||||
}
|
||||
|
||||
|
||||
def allowed_fields() -> frozenset[str]:
|
||||
"""条件可引用字段 = 物化列白名单 并入 注册表因子 (虚拟/自定义/复合)。
|
||||
|
||||
因子列在历史路径 (compute_signals) 由 materialize_factor_columns 复用
|
||||
评分物化管线补算; 盘中单日快照无滚动窗口, 依赖因子的信号被 inject 以
|
||||
缺列告警跳过 (与日期偏移条件同样的优雅降级)。
|
||||
"""
|
||||
from app.factors.registry import all_factors
|
||||
|
||||
return frozenset(ALLOWED_FIELDS | {spec.id for spec in all_factors()})
|
||||
|
||||
|
||||
def materialize_factor_columns(
|
||||
df: pl.DataFrame,
|
||||
exprs: dict[str, pl.Expr],
|
||||
needed: set[str] | None = None,
|
||||
) -> pl.DataFrame:
|
||||
"""把信号表达式引用、且 df 缺失的注册表因子列补算出来。
|
||||
|
||||
复用评分物化路径 (materialize_scoring_columns) — 与检验/评分同一条计算
|
||||
逻辑, 不引入第二套实现。非注册表列不在此处理 (缺列仍由 inject 告警跳过)。
|
||||
"""
|
||||
if df.is_empty() or not exprs:
|
||||
return df
|
||||
cols = set(df.columns)
|
||||
missing: set[str] = set()
|
||||
for name, roots in expression_dependencies(exprs).items():
|
||||
if needed is not None and name not in needed:
|
||||
continue
|
||||
missing.update(root for root in roots if root not in cols)
|
||||
if not missing:
|
||||
return df
|
||||
from app.factors.registry import all_factors
|
||||
|
||||
factor_ids = {spec.id for spec in all_factors()}
|
||||
to_compute = missing & factor_ids
|
||||
if not to_compute:
|
||||
return df
|
||||
from app.strategy.scoring import materialize_scoring_columns
|
||||
|
||||
return materialize_scoring_columns(df, sorted(to_compute))
|
||||
|
||||
|
||||
# ── 持久化（镜像 strategy/config.py 的写法）──────────────
|
||||
def _dir(data_dir: Path) -> Path:
|
||||
d = data_dir / "user_data" / "custom_signals"
|
||||
@@ -133,9 +176,10 @@ def _parse_right(right: str) -> tuple[str, object]:
|
||||
return ("const", float(right))
|
||||
if not isinstance(right, str):
|
||||
raise ValueError(f"非法右值: {right!r}")
|
||||
allowed = allowed_fields()
|
||||
if right.startswith("field:"):
|
||||
col = right[len("field:"):]
|
||||
if col not in ALLOWED_FIELDS:
|
||||
if col not in allowed:
|
||||
raise ValueError(f"右值字段不在白名单: {col}")
|
||||
return ("field", col)
|
||||
# 纯数字
|
||||
@@ -144,7 +188,7 @@ def _parse_right(right: str) -> tuple[str, object]:
|
||||
except ValueError:
|
||||
pass
|
||||
# 裸字段名 — 兜底容错, 仍受白名单约束
|
||||
if right in ALLOWED_FIELDS:
|
||||
if right in allowed:
|
||||
return ("field", right)
|
||||
raise ValueError(f"非法右值（应为 field:xxx 或数字）: {right!r}")
|
||||
|
||||
@@ -167,7 +211,7 @@ def validate(sig: dict) -> None:
|
||||
if not isinstance(c, dict):
|
||||
raise ValueError(f"第 {i+1} 个条件格式错误")
|
||||
left = c.get("left", "")
|
||||
if left not in ALLOWED_FIELDS:
|
||||
if left not in allowed_fields():
|
||||
raise ValueError(f"第 {i+1} 个条件: 字段 {left!r} 不在白名单")
|
||||
if c.get("op") not in OPS:
|
||||
raise ValueError(f"第 {i+1} 个条件: 运算符 {c.get('op')!r} 非法")
|
||||
|
||||
@@ -32,8 +32,12 @@ _FENCED_JSON_RE = re.compile(r"```(?:json)?\s*\n?(.*?)```", re.DOTALL)
|
||||
|
||||
|
||||
def _format_fields() -> str:
|
||||
"""按类别格式化白名单字段（key(中文标签)），供 LLM 参考。"""
|
||||
allowed = custom_signals.ALLOWED_FIELDS
|
||||
"""按类别格式化白名单字段(key(中文标签)), 供 LLM 参考.
|
||||
|
||||
行情/指标类物理列之后追加注册表因子, 分组与 /api/custom-signals/options
|
||||
的 factor 分组一致: 因子是预计算因子值, 同样可作为条件字段比较.
|
||||
"""
|
||||
allowed = custom_signals.allowed_fields()
|
||||
lines: list[str] = []
|
||||
quote = sorted(f for f in _QUOTE_FIELDS if f in allowed)
|
||||
lines.append(
|
||||
@@ -46,6 +50,18 @@ def _format_fields() -> str:
|
||||
f"{label}: "
|
||||
+ ", ".join(f"{f}({ENRICHED_COLUMNS.get(f, f)})" for f in fields)
|
||||
)
|
||||
from app.factors.registry import all_factors
|
||||
|
||||
factor_groups: dict[str, list[str]] = {}
|
||||
for spec in all_factors():
|
||||
if spec.id in custom_signals.ALLOWED_FIELDS:
|
||||
continue # 已作为物理列出现在清单里
|
||||
label = spec.label
|
||||
if spec.asset_types == frozenset({"stock"}):
|
||||
label += "·仅股票"
|
||||
factor_groups.setdefault(spec.group, []).append(f"{spec.id}({label})")
|
||||
for group, items in sorted(factor_groups.items()):
|
||||
lines.append(f"因子·{group}: " + ", ".join(sorted(items)))
|
||||
return "\n".join(lines)
|
||||
|
||||
|
||||
@@ -53,6 +69,7 @@ _SYSTEM_TEMPLATE = """你是A股量化信号设计专家。用户会描述一个
|
||||
|
||||
可用字段（白名单，只能使用以下字段，禁止自造或使用白名单之外的字段）：
|
||||
{fields}
|
||||
其中「因子·」开头的行是平台预计算的因子值（动量/波动/量价等衍生特征），可直接比较数值构造条件。
|
||||
|
||||
运算符（op）：> >= < <= == !=
|
||||
|
||||
@@ -132,8 +149,8 @@ def _normalize_condition(c: object) -> dict:
|
||||
if not isinstance(right, str) or not right.strip():
|
||||
raise ValueError(f"右值非法: {right!r}")
|
||||
right = right.strip()
|
||||
# 兜底: AI 偶尔漏写 field: 前缀的裸字段名, 补全为规范形式
|
||||
if not right.startswith("field:") and right in custom_signals.ALLOWED_FIELDS:
|
||||
# 兜底: AI 偶尔漏写 field: 前缀的裸字段名, 补全为规范形式 (含因子字段)
|
||||
if not right.startswith("field:") and right in custom_signals.allowed_fields():
|
||||
right = f"field:{right}"
|
||||
return {
|
||||
"left": str(left),
|
||||
|
||||
+126
-56
@@ -6,61 +6,22 @@ from typing import Any
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.factors.registry import (
|
||||
factor_dependencies as _registry_factor_dependencies,
|
||||
)
|
||||
from app.factors.registry import get_factor as _registry_get_factor
|
||||
from app.factors.registry import scoring_warmups as _registry_scoring_warmups
|
||||
from app.factors.registry import virtual_dependencies as _registry_virtual_dependencies
|
||||
|
||||
SCORING_DIRECTION_HIGH = "high"
|
||||
SCORING_DIRECTION_LOW = "low"
|
||||
SCORING_DIRECTIONS = frozenset({SCORING_DIRECTION_HIGH, SCORING_DIRECTION_LOW})
|
||||
|
||||
VIRTUAL_SCORING_DEPENDENCIES: dict[str, frozenset[str]] = {
|
||||
**{
|
||||
f"ma{period}_bias": frozenset({"close", f"ma{period}"})
|
||||
for period in (5, 10, 20, 30, 60)
|
||||
},
|
||||
**{
|
||||
f"ema{period}_bias": frozenset({"close", f"ema{period}"})
|
||||
for period in (5, 10, 20, 30, 60)
|
||||
},
|
||||
"macd_dif_pct": frozenset({"close", "macd_dif"}),
|
||||
"macd_dea_pct": frozenset({"close", "macd_dea"}),
|
||||
"macd_hist_pct": frozenset({"close", "macd_hist"}),
|
||||
"boll_position": frozenset({"close", "boll_upper", "boll_lower"}),
|
||||
"atr_pct": frozenset({"close", "atr_14"}),
|
||||
"boll_width": frozenset({"ma20", "boll_upper", "boll_lower"}),
|
||||
"vol_ratio_10d": frozenset({"volume"}),
|
||||
"vol_trend_5_10": frozenset({"vol_ma5", "vol_ma10"}),
|
||||
"turnover_ratio_5d": frozenset({"turnover_rate"}),
|
||||
"log_amount": frozenset({"amount"}),
|
||||
"amount_ratio_5d": frozenset({"amount"}),
|
||||
"gap_return": frozenset({"open", "prev_close"}),
|
||||
"intraday_return": frozenset({"open", "close"}),
|
||||
"close_position": frozenset({"high", "low", "close"}),
|
||||
"distance_to_high_60d": frozenset({"close", "high_60d"}),
|
||||
"distance_from_low_60d": frozenset({"close", "low_60d"}),
|
||||
"max_ret_20d": frozenset({"close"}),
|
||||
"ret_skew_20d": frozenset({"close"}),
|
||||
"up_days_20d": frozenset({"close"}),
|
||||
"amihud_20d": frozenset({"close", "amount"}),
|
||||
"turnover_z_60d": frozenset({"turnover_rate"}),
|
||||
"vol_price_corr_20d": frozenset({"close", "volume"}),
|
||||
"vwap_bias": frozenset({"close", "volume", "amount"}),
|
||||
"vol_trend_5_60": frozenset({"volume"}),
|
||||
"limit_up_count_20d": frozenset({"consecutive_limit_ups"}),
|
||||
"limit_up_count_60d": frozenset({"consecutive_limit_ups"}),
|
||||
}
|
||||
# P1 起依赖声明与预热窗口的单一权威来源为 app/factors/registry.py;
|
||||
# 本常量为兼容别名, 键集合与历史版本逐项一致 (见 tests/test_factor_registry.py 快照测试)。
|
||||
VIRTUAL_SCORING_DEPENDENCIES: dict[str, frozenset[str]] = dict(_registry_virtual_dependencies())
|
||||
|
||||
_ROLLING_SCORING_WARMUP: dict[str, int] = {
|
||||
"vol_ratio_10d": 11,
|
||||
"turnover_ratio_5d": 6,
|
||||
"amount_ratio_5d": 6,
|
||||
"max_ret_20d": 21,
|
||||
"ret_skew_20d": 21,
|
||||
"up_days_20d": 21,
|
||||
"amihud_20d": 21,
|
||||
"turnover_z_60d": 61,
|
||||
"vol_price_corr_20d": 21,
|
||||
"vol_trend_5_60": 60,
|
||||
"limit_up_count_20d": 21,
|
||||
"limit_up_count_60d": 61,
|
||||
}
|
||||
_ROLLING_SCORING_WARMUP: dict[str, int] = dict(_registry_scoring_warmups())
|
||||
|
||||
|
||||
def effective_scoring(
|
||||
@@ -89,27 +50,61 @@ def effective_scoring_directions(overrides: Mapping[str, Any] | None) -> dict[st
|
||||
|
||||
|
||||
def scoring_warmup_bars(scoring: Mapping[str, Any]) -> int:
|
||||
return max(
|
||||
(_ROLLING_SCORING_WARMUP.get(str(name), 1) for name, weight in scoring.items() if weight),
|
||||
default=1,
|
||||
)
|
||||
warmups: list[int] = [
|
||||
_ROLLING_SCORING_WARMUP.get(str(name), 1)
|
||||
for name, weight in scoring.items()
|
||||
if weight
|
||||
]
|
||||
# composite/custom 因子的预热来自注册表 (P3)
|
||||
for name, weight in scoring.items():
|
||||
if not weight:
|
||||
continue
|
||||
spec = _registry_get_factor(str(name))
|
||||
if spec is not None and spec.kind in ("custom", "composite"):
|
||||
warmups.append(spec.warmup_bars)
|
||||
return max(warmups, default=1)
|
||||
|
||||
|
||||
def scoring_dependencies(scoring: Mapping[str, Any]) -> set[str]:
|
||||
"""把受控虚拟评分字段展开为实际数据依赖。"""
|
||||
"""把受控虚拟评分字段展开为实际数据依赖 (含 composite/custom 递归展开)。"""
|
||||
dependencies: set[str] = set()
|
||||
for name, weight in scoring.items():
|
||||
if not weight:
|
||||
continue
|
||||
dependencies.update(VIRTUAL_SCORING_DEPENDENCIES.get(str(name), {str(name)}))
|
||||
dependencies.update(_registry_factor_dependencies([str(name)]))
|
||||
return dependencies
|
||||
|
||||
|
||||
def _composite_value_expr(available: set[str], name: str) -> pl.Expr | None:
|
||||
"""复合因子值 = Σ w_i * 截面 zscore(成员值); 成员可为已物化列或虚拟因子。"""
|
||||
spec = _registry_get_factor(name)
|
||||
if spec is None or not spec.components:
|
||||
return None
|
||||
total: pl.Expr | None = None
|
||||
for member_id, weight in spec.components:
|
||||
member_expr = (
|
||||
pl.col(member_id)
|
||||
if member_id in available
|
||||
else scoring_value_expr(available, member_id)
|
||||
)
|
||||
if member_expr is None:
|
||||
return None
|
||||
mean = member_expr.mean().over("date")
|
||||
std = member_expr.std().over("date")
|
||||
piece = pl.when(std > 0).then((member_expr - mean) / std).otherwise(None) * weight
|
||||
total = piece if total is None else total + piece
|
||||
return total
|
||||
|
||||
|
||||
def scoring_value_expr(columns: Collection[str], name: str) -> pl.Expr | None:
|
||||
"""返回评分值表达式；依赖不完整时返回 None。"""
|
||||
available = set(columns)
|
||||
if name in available:
|
||||
return pl.col(name)
|
||||
# composite 在 VIRTUAL 字典门控之前分派 (依赖经注册表递归展开) —— P3
|
||||
spec = _registry_get_factor(name)
|
||||
if spec is not None and spec.kind == "composite":
|
||||
return _composite_value_expr(available, name)
|
||||
dependencies = VIRTUAL_SCORING_DEPENDENCIES.get(name)
|
||||
if dependencies is None or not dependencies.issubset(available):
|
||||
return None
|
||||
@@ -202,6 +197,66 @@ def scoring_value_expr(columns: Collection[str], name: str) -> pl.Expr | None:
|
||||
window = 20 if name == "limit_up_count_20d" else 60
|
||||
hit = (pl.col("consecutive_limit_ups").fill_null(0) > 0).cast(pl.Float64)
|
||||
return hit.rolling_sum(window, min_samples=window).over("symbol")
|
||||
# ── 扩充批次 (2026-09-05): 全部滚动窗口默认 min_samples=窗口长 (fail-closed) ──
|
||||
if name == "log_float_mv":
|
||||
# 换手率 = 成交量/流通股本 → 股本 = volume/turnover_rate, 市值 = close x 股本
|
||||
return (
|
||||
pl.when((pl.col("turnover_rate") > 0) & (pl.col("volume") > 0))
|
||||
.then((pl.col("close") * pl.col("volume") / pl.col("turnover_rate")).log())
|
||||
.otherwise(None)
|
||||
)
|
||||
if name == "momentum_120d":
|
||||
return _relative(
|
||||
pl.col("close"),
|
||||
pl.col("close").shift(120),
|
||||
).over("symbol")
|
||||
if name == "mom_accel_20_60":
|
||||
return pl.col("momentum_20d") - pl.col("momentum_60d")
|
||||
if name == "rsi_14_delta_5d":
|
||||
return pl.col("rsi_14") - pl.col("rsi_14").shift(5).over("symbol")
|
||||
if name == "overnight_ret_20d":
|
||||
overnight = _relative(pl.col("open"), pl.col("prev_close"))
|
||||
return overnight.rolling_sum(20, min_samples=20).over("symbol")
|
||||
if name == "intraday_ret_20d":
|
||||
intraday = _relative(pl.col("close"), pl.col("open"))
|
||||
return intraday.rolling_sum(20, min_samples=20).over("symbol")
|
||||
if name == "downside_vol_20d":
|
||||
downside = (_daily_change_expr().clip(upper_bound=0.0) ** 2)
|
||||
return downside.rolling_mean(20, min_samples=20).sqrt().over("symbol")
|
||||
if name == "vol_regime_5_60":
|
||||
change = _daily_change_expr()
|
||||
fast = change.rolling_std(5, min_samples=5)
|
||||
slow = change.rolling_std(60, min_samples=60)
|
||||
return _ratio(fast, slow).over("symbol")
|
||||
if name == "amplitude_trend_20_60":
|
||||
fast = pl.col("amplitude").rolling_mean(20, min_samples=20)
|
||||
slow = pl.col("amplitude").rolling_mean(60, min_samples=60)
|
||||
return _relative(fast, slow).over("symbol")
|
||||
if name == "obv_trend_20d":
|
||||
change = _daily_change_expr()
|
||||
signed = change.sign() * pl.col("volume")
|
||||
total = signed.rolling_sum(20, min_samples=20)
|
||||
scale = pl.col("volume").rolling_mean(20, min_samples=20) * 20.0
|
||||
return _ratio(total, scale).over("symbol")
|
||||
if name == "amount_mean_20d":
|
||||
return (pl.col("amount") / 1e8).rolling_mean(20, min_samples=20).over("symbol")
|
||||
if name == "turnover_mean_20d":
|
||||
return pl.col("turnover_rate").rolling_mean(20, min_samples=20).over("symbol")
|
||||
if name == "turnover_std_20d":
|
||||
mean = pl.col("turnover_rate").rolling_mean(20, min_samples=20)
|
||||
std = pl.col("turnover_rate").rolling_std(20, min_samples=20)
|
||||
return _ratio(std, mean).over("symbol")
|
||||
if name == "position_240d":
|
||||
high = pl.col("close").rolling_max(240, min_samples=240)
|
||||
low = pl.col("close").rolling_min(240, min_samples=240)
|
||||
return _ratio(pl.col("close") - low, high - low).over("symbol")
|
||||
if name == "distance_to_high_240d":
|
||||
return _relative(
|
||||
pl.col("close"),
|
||||
pl.col("close").rolling_max(240, min_samples=240),
|
||||
).over("symbol")
|
||||
if name == "kdj_kd_diff":
|
||||
return pl.col("kdj_k") - pl.col("kdj_d")
|
||||
return None
|
||||
|
||||
|
||||
@@ -233,6 +288,21 @@ def materialize_scoring_columns(
|
||||
frame: pl.DataFrame,
|
||||
names: Collection[str],
|
||||
) -> pl.DataFrame:
|
||||
# custom (DSL) 因子先物化: frame_transform 可能需要多阶段临时列 (嵌套窗口规避),
|
||||
# 与单表达式路径不同, 必须整体走帧变换 —— 与检验/试算共用同一条计算路径 (P3)。
|
||||
from app.factors.dsl import FACTOR_COLUMN, compile_formula_cached
|
||||
|
||||
for name in names:
|
||||
spec = _registry_get_factor(str(name))
|
||||
if spec is None or spec.kind != "custom" or name in frame.columns:
|
||||
continue
|
||||
compiled = compile_formula_cached(spec.formula_text)
|
||||
if compiled.frame_transform is None:
|
||||
continue
|
||||
transformed = compiled.frame_transform(frame)
|
||||
if transformed is None:
|
||||
continue
|
||||
frame = transformed.with_columns(pl.col(FACTOR_COLUMN).alias(str(name))).drop(FACTOR_COLUMN)
|
||||
expressions = [
|
||||
expression.alias(name)
|
||||
for name in names
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
[project]
|
||||
name = "tickflow-stock-panel-backend"
|
||||
version = "0.2.2"
|
||||
version = "0.2.3"
|
||||
description = "A 股选股 + 监控 + 回测面板 — TickFlow 适配"
|
||||
requires-python = ">=3.11"
|
||||
license = { text = "MIT" }
|
||||
|
||||
@@ -636,3 +636,57 @@ def test_matrix_cache_preserves_trades_daily_equity_and_core_stats():
|
||||
assert cached.stats[name] == uncached.stats[name]
|
||||
assert cached_again.stats[name] == uncached.stats[name]
|
||||
assert cached_again.stats["matrix_compute_cache"]["hits"] > 0
|
||||
|
||||
|
||||
def test_apply_score_captures_factor_snapshot_for_candidates():
|
||||
panel = pl.DataFrame({
|
||||
"symbol": ["A", "B", "C"],
|
||||
"date": [date(2024, 1, 1)] * 3,
|
||||
"factor": [10.0, 20.0, 1000.0],
|
||||
})
|
||||
universe = pl.Series([True, True, False], dtype=pl.Boolean)
|
||||
strategy = SimpleNamespace(meta={"scoring": {"factor": 1.0}, "order_by": "score", "descending": True})
|
||||
|
||||
snapshot: dict = {}
|
||||
StrategyBacktestService._apply_score(panel, strategy, None, universe_mask=universe, factor_snapshot=snapshot)
|
||||
frame = snapshot["frame"]
|
||||
assert frame.columns == ["symbol", "date", "factor"]
|
||||
assert frame["symbol"].to_list() == ["A", "B"] # 非候选行不入快照
|
||||
|
||||
|
||||
def test_full_mode_reports_factor_attribution():
|
||||
start = date(2024, 1, 1)
|
||||
panel = pl.DataFrame([
|
||||
{"symbol": "A", "name": "A", "date": start, "open": 10.0, "high": 10.0, "low": 10.0, "close": 10.0, "volume": 1, "amount": 1000.0, "signal_limit_up": False, "signal_limit_down": False, "factor": 5.0},
|
||||
{"symbol": "A", "name": "A", "date": start + timedelta(days=1), "open": 11.0, "high": 11.0, "low": 11.0, "close": 11.0, "volume": 1, "amount": 0.0, "signal_limit_up": False, "signal_limit_down": False, "factor": 5.0},
|
||||
{"symbol": "A", "name": "A", "date": start + timedelta(days=2), "open": 20.0, "high": 20.0, "low": 20.0, "close": 20.0, "volume": 1, "amount": 1000.0, "signal_limit_up": False, "signal_limit_down": False, "factor": 5.0},
|
||||
]).sort(["symbol", "date"])
|
||||
|
||||
engine = BacktestEngine(repo=None) # type: ignore[arg-type]
|
||||
engine.load_panel_for_backtest = lambda symbols, s, e, plan, asset_type="stock": panel # type: ignore[method-assign]
|
||||
strategy = _strategy(
|
||||
filter_fn=lambda df, params: pl.col("date") == start,
|
||||
max_hold_days=1,
|
||||
)
|
||||
strategy.meta["scoring"] = {"factor": 1.0}
|
||||
service = StrategyBacktestService(engine=engine, strategy_engine=_StrategyEngineStub(strategy))
|
||||
|
||||
result = service.run(StrategyBacktestConfig(
|
||||
strategy_id="test",
|
||||
symbols=None,
|
||||
start=start,
|
||||
end=start,
|
||||
mode="full",
|
||||
matching="open_t+1",
|
||||
fees_pct=0,
|
||||
slippage_bps=0,
|
||||
holding_days=1,
|
||||
))
|
||||
|
||||
assert result.error is None
|
||||
assert result.stats["n_trades"] == 1
|
||||
attribution = result.factor_attribution
|
||||
assert attribution is not None
|
||||
assert attribution["n_win"] == 1 and attribution["n_lose"] == 0
|
||||
assert attribution["factors"][0]["factor"] == "factor"
|
||||
assert attribution["factors"][0]["win_mean"] == 5.0
|
||||
|
||||
@@ -465,3 +465,90 @@ def test_engine_abnormal_down_direction_event_type() -> None:
|
||||
events = engine.evaluate_abnormal([_row("600000.SH", ("3d", -0.16))], now=1006.0)
|
||||
assert len(events) == 1
|
||||
assert events[0]["type"] == "abnormal_down"
|
||||
|
||||
|
||||
# ── 板块基准路由: 科创板→科创50, 创业板→创业板综指 (交易所官方对应指数) ──
|
||||
|
||||
def test_attach_deviation_columns_board_routing(tmp_path) -> None:
|
||||
"""科创板股票减科创50、创业板股票减创业板综指, 不再共用交易所 A 指。
|
||||
|
||||
构造 68/30 前缀股票 close 与各自官方基准同序列 → 偏离恰为 0。
|
||||
"""
|
||||
days = [date(2026, 8, 13), date(2026, 8, 14), date(2026, 8, 15), date(2026, 8, 18)]
|
||||
rows = [("000688.SH", d, 100.0 + i) for i, d in enumerate(days)] # 科创50
|
||||
rows += [("399102.SZ", d, 200.0 + 5 * i) for i, d in enumerate(days)] # 创业板综指
|
||||
_write_index_daily(tmp_path, rows)
|
||||
|
||||
stock = pl.DataFrame(
|
||||
{
|
||||
"symbol": ["688981.SH"] * 4 + ["300750.SZ"] * 4,
|
||||
"date": days * 2,
|
||||
"close": [100.0 + i for i in range(4)] + [200.0 + 5 * i for i in range(4)],
|
||||
}
|
||||
)
|
||||
out = attach_deviation_columns(stock, tmp_path)
|
||||
star = out.filter(pl.col("symbol") == "688981.SH").sort("date")["deviate_3d"][-1]
|
||||
gem = out.filter(pl.col("symbol") == "300750.SZ").sort("date")["deviate_3d"][-1]
|
||||
assert abs(star - 0.0) < 1e-9
|
||||
assert abs(gem - 0.0) < 1e-9
|
||||
|
||||
|
||||
def test_attach_deviation_columns_star_fallback(tmp_path) -> None:
|
||||
"""科创50 数据缺失 → 回退上证A指, 偏离列不整体缺失。"""
|
||||
days = [date(2026, 8, 13), date(2026, 8, 14), date(2026, 8, 15), date(2026, 8, 18)]
|
||||
_write_index_daily(tmp_path, [("000002.SH", d, 10.0 + i) for i, d in enumerate(days)])
|
||||
|
||||
stock = pl.DataFrame(
|
||||
{
|
||||
"symbol": ["688981.SH"] * 4,
|
||||
"date": days,
|
||||
"close": [10.0 + i for i in range(4)],
|
||||
}
|
||||
)
|
||||
out = attach_deviation_columns(stock, tmp_path)
|
||||
dev = out.sort("date")["deviate_3d"][-1]
|
||||
assert dev is not None
|
||||
assert abs(dev - 0.0) < 1e-9
|
||||
|
||||
|
||||
def test_benchmark_momentum_today_gem_key(tmp_path) -> None:
|
||||
"""盘中外推按板块键产出: 创业板综指行情只喂 GEM 键的实时涨跌。"""
|
||||
_write_index_daily(tmp_path, [("399102.SZ", d, 10.0 + i) for i, d in enumerate(_BENCH_DAYS)])
|
||||
quotes = pl.DataFrame({"symbol": ["399102.SZ"], "change_pct": [10.0]}) # 百分数: +10%
|
||||
|
||||
out = benchmark_momentum_today(tmp_path, quotes)
|
||||
gem = out.filter(pl.col("bench_key") == "GEM")
|
||||
assert gem.height == 1
|
||||
assert abs(gem["bench_mom3d"][0] - (15.0 * 1.10 / 13 - 1)) < 1e-9
|
||||
# 其他键 (如 SH) 候选不在实时缓存 → rt=0, 用昨收外推
|
||||
sh = out.filter(pl.col("bench_key") == "SH")
|
||||
assert sh.height == 1
|
||||
assert abs(sh["bench_mom3d"][0] - (15.0 / 13 - 1)) < 1e-9
|
||||
|
||||
|
||||
def test_build_overview_rt_overlay_uses_board_benchmark() -> None:
|
||||
"""实时叠加按板块基准: 创业板股减创业板综指今日涨跌, 不再全市场混均值。"""
|
||||
with _hist_cache_lock:
|
||||
_hist_cache.clear()
|
||||
|
||||
class _GemQuotes:
|
||||
def get_index_quotes(self):
|
||||
# 百分数口径: 创业板综指 +2%, 其余基准无行情 → 0
|
||||
return pl.DataFrame({"symbol": ["399102.SZ"], "change_pct": [2.0]})
|
||||
|
||||
df = pl.DataFrame(
|
||||
{
|
||||
"symbol": ["300001.SZ", "600000.SH"],
|
||||
"name": ["创业板股", "沪主板股"],
|
||||
"close": [10.0, 20.0],
|
||||
"change_pct": [0.10, 0.10],
|
||||
"deviate_3d": [0.25, 0.25],
|
||||
"deviate_10d": [None, None],
|
||||
"deviate_30d": [None, None],
|
||||
}
|
||||
)
|
||||
result = build_overview(_FakeRepo(df), _GemQuotes(), min_closeness=0.5)
|
||||
by_symbol = {r["symbol"]: r for r in result["rows"]}
|
||||
# 创业板: 0.25 + (0.10 - 0.02) = 0.33; 沪主板无对应行情: 0.25 + 0.10 = 0.35
|
||||
assert abs(by_symbol["300001.SZ"]["windows"]["3d"]["value"] - 0.33) < 1e-9
|
||||
assert abs(by_symbol["600000.SH"]["windows"]["3d"]["value"] - 0.35) < 1e-9
|
||||
|
||||
@@ -0,0 +1,214 @@
|
||||
"""自动挖掘 L1 筛选 (auto_mining) 与 /api/backtest/mining/auto 契约测试。"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date, timedelta
|
||||
from types import SimpleNamespace
|
||||
from typing import Any, ClassVar
|
||||
|
||||
import pytest
|
||||
|
||||
from app.services import auto_mining
|
||||
from app.services.auto_mining import (
|
||||
SCREEN_GATES,
|
||||
classify_factor,
|
||||
screen_all_factors,
|
||||
)
|
||||
|
||||
# ── classify_factor: 门槛分支 ──
|
||||
|
||||
|
||||
def _item(**overrides: Any) -> dict[str, Any]:
|
||||
base = {"error": None, "ic_mean": 0.05, "ir": 0.4, "t_newey_west": 2.5, "q_value": 0.05}
|
||||
base.update(overrides)
|
||||
return base
|
||||
|
||||
|
||||
def test_classify_balanced_gate_branches() -> None:
|
||||
gate = SCREEN_GATES["balanced"]
|
||||
assert classify_factor(_item(), gate) is None
|
||||
assert "计算失败" in classify_factor(_item(error="boom"), gate)
|
||||
assert "样本不足" in classify_factor(_item(ic_mean=None), gate)
|
||||
assert "预测力弱" in classify_factor(_item(ic_mean=0.01), gate)
|
||||
assert "稳定度低" in classify_factor(_item(ic_mean=0.05, ir=0.1), gate)
|
||||
assert "NW t" in classify_factor(_item(t_newey_west=None), gate)
|
||||
assert "不显著" in classify_factor(_item(t_newey_west=1.0), gate)
|
||||
assert "多重检验" in classify_factor(_item(q_value=0.5), gate)
|
||||
# q 缺失按通过 (探索档小样本口径)
|
||||
assert classify_factor(_item(q_value=None), gate) is None
|
||||
|
||||
|
||||
def test_classify_profile_gates_tighten() -> None:
|
||||
item = _item(ic_mean=-0.025, ir=-0.35, t_newey_west=-2.2, q_value=0.08)
|
||||
# 负值同样达标 (方向反向), 严格档收紧后不达标
|
||||
assert classify_factor(item, SCREEN_GATES["balanced"]) is None
|
||||
assert classify_factor(item, SCREEN_GATES["strict"]) is not None
|
||||
# 探索档最宽
|
||||
weak = _item(ic_mean=0.021, ir=0.16, t_newey_west=1.6, q_value=0.18)
|
||||
assert classify_factor(weak, SCREEN_GATES["exploratory"]) is None
|
||||
assert classify_factor(weak, SCREEN_GATES["balanced"]) is not None
|
||||
|
||||
|
||||
# ── screen_all_factors: 池构造/排序/截断/清洗 ──
|
||||
|
||||
|
||||
class _StubService:
|
||||
calls: ClassVar[list[Any]] = []
|
||||
results: ClassVar[list[Any]] = []
|
||||
|
||||
def run_batch(self, config: Any) -> Any:
|
||||
_StubService.calls.append(config)
|
||||
return SimpleNamespace(results=_StubService.results)
|
||||
|
||||
|
||||
def _batch_item(name: str, **overrides: Any) -> SimpleNamespace:
|
||||
base = {
|
||||
"factor_name": name, "label": f"label_{name}", "group": "g",
|
||||
"ic_mean": 0.05, "ir": 0.4, "t_newey_west": 2.5, "q_value": 0.05,
|
||||
"error": None,
|
||||
}
|
||||
base.update(overrides)
|
||||
return SimpleNamespace(**base)
|
||||
|
||||
|
||||
@pytest.fixture()
|
||||
def _stub_batch(monkeypatch: pytest.MonkeyPatch) -> type[_StubService]:
|
||||
_StubService.calls = []
|
||||
monkeypatch.setattr(auto_mining, "FactorBacktestService", lambda engine: _StubService())
|
||||
return _StubService
|
||||
|
||||
|
||||
def test_screen_pool_order_truncation_and_reasons(monkeypatch: pytest.MonkeyPatch, _stub_batch: type[_StubService]) -> None:
|
||||
monkeypatch.setattr(auto_mining, "factor_columns_view", lambda: [
|
||||
{"id": name, "asset_types": ["stock"]} for name in ("a", "b", "c", "d", "e", "f")
|
||||
])
|
||||
items = [
|
||||
_batch_item("a", ic_mean=0.10, ir=0.8), # |ic|*|ir|=0.08 → 第 1
|
||||
_batch_item("b", ic_mean=0.05, ir=0.4), # 0.02 → 第 2
|
||||
_batch_item("c", ic_mean=-0.06, ir=-0.5), # 0.03 → 第 3 (负 IC 反向)
|
||||
_batch_item("d", ic_mean=0.01), # 预测力弱
|
||||
_batch_item("e", ir=0.1), # 稳定度低
|
||||
_batch_item("f", t_newey_west=float("nan")), # NaN 清洗 → 样本不足
|
||||
]
|
||||
_stub_batch.results = items
|
||||
summary = screen_all_factors(
|
||||
object(), asset_type="stock",
|
||||
start=date.today() - timedelta(days=800),
|
||||
end=date.today(), profile="balanced",
|
||||
max_factors=2,
|
||||
)
|
||||
assert summary["pool"] == ["a", "c"] # 0.08 > 0.03 > 0.02, 截断到 2
|
||||
assert summary["pool_truncated"] is True
|
||||
assert summary["n_qualified"] == 3
|
||||
assert summary["n_total"] == len(summary["qualified"]) + len(summary["failed"])
|
||||
assert {row["factor_name"]: row["direction"] for row in summary["qualified"]}["c"] == -1
|
||||
# NaN 指标被清洗为 None, 归入样本不足而非写入非法 JSON
|
||||
row_f = next(row for row in summary["failed"] if row["factor_name"] == "f")
|
||||
assert row_f["t"] is None and "样本不足" in row_f["reason"]
|
||||
counts = summary["reason_counts"]
|
||||
assert counts["预测力弱"] == 1
|
||||
assert counts["稳定度低"] == 1
|
||||
assert any(key.startswith("样本不足") for key in counts)
|
||||
assert sum(counts.values()) == 3
|
||||
|
||||
|
||||
def test_screen_window_capped_and_daily_rebalance(monkeypatch: pytest.MonkeyPatch, _stub_batch: type[_StubService]) -> None:
|
||||
_stub_batch.results = []
|
||||
start = date.today() - timedelta(days=1000)
|
||||
screen_all_factors(
|
||||
object(), asset_type="stock", start=start, end=date.today(),
|
||||
profile="exploratory",
|
||||
)
|
||||
config = _stub_batch.calls[0]
|
||||
assert config.rebalance == "daily"
|
||||
assert (config.end - config.start).days <= auto_mining.SCREEN_WINDOW_DAYS
|
||||
# 因子清单来自注册表 (动态视图, 数量与内置一致量级)
|
||||
assert len(config.factor_names) >= 50
|
||||
|
||||
|
||||
# ── API 契约: /api/backtest/mining/auto ──
|
||||
|
||||
|
||||
class _FakeManager:
|
||||
def __init__(self, store: Any) -> None:
|
||||
self.store = store
|
||||
self.start_calls: list[dict[str, Any]] = []
|
||||
|
||||
def start(self, request, fingerprint, force=False, source="manual", run_id=None):
|
||||
self.start_calls.append({"request": request, "fingerprint": fingerprint,
|
||||
"force": force, "source": source})
|
||||
return self.store.create(request, fingerprint)
|
||||
|
||||
|
||||
@pytest.fixture()
|
||||
def client(monkeypatch: pytest.MonkeyPatch, tmp_path: Any) -> Any:
|
||||
from fastapi import FastAPI
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from app.api import mining as mining_api
|
||||
from app.services.mining_jobs import MiningRunStore
|
||||
|
||||
store = MiningRunStore(tmp_path / "runs")
|
||||
manager = _FakeManager(store)
|
||||
|
||||
app = FastAPI()
|
||||
app.include_router(mining_api.router)
|
||||
app.state.repo = SimpleNamespace(store=SimpleNamespace(data_dir=tmp_path))
|
||||
app.state.mining_manager = manager
|
||||
|
||||
monkeypatch.setattr(mining_api, "require_mining_availability", lambda *a, **k: None)
|
||||
monkeypatch.setattr(mining_api, "enriched_partition_dates", lambda *a, **k: ["2026-08-31"])
|
||||
monkeypatch.setattr(mining_api, "build_data_fingerprint", lambda *a, **k: {"fp": 1})
|
||||
return SimpleNamespace(client=TestClient(app), manager=manager, store=store)
|
||||
|
||||
|
||||
def _screening(pool: list[str]) -> dict[str, Any]:
|
||||
return {
|
||||
"profile": "balanced", "gate": {"min_abs_ic": 0.02}, "n_total": 61,
|
||||
"n_qualified": len(pool), "pool": pool, "pool_truncated": False,
|
||||
"qualified": [], "failed": [], "reason_counts": {},
|
||||
"screen_window": {"start": "2025-08-31", "end": "2026-08-31"},
|
||||
"elapsed_ms": 1.0,
|
||||
}
|
||||
|
||||
|
||||
def test_auto_start_contract_with_pool(client: Any, monkeypatch: pytest.MonkeyPatch) -> None:
|
||||
monkeypatch.setattr(
|
||||
auto_mining, "screen_all_factors",
|
||||
lambda *a, **k: _screening(["f1", "f2"]),
|
||||
)
|
||||
response = client.client.post("/api/backtest/mining/auto", json={"asset_type": "stock"})
|
||||
assert response.status_code == 200
|
||||
body = response.json()
|
||||
assert body["started"] is True
|
||||
assert body["run"]["status"] == "queued"
|
||||
call = client.manager.start_calls[0]
|
||||
assert call["source"] == "auto"
|
||||
request = call["request"]
|
||||
assert request["factor_names"] == ["f1", "f2"]
|
||||
assert request["strategy_ids"] == []
|
||||
assert request["auto"] is True
|
||||
assert request["auto_screening"]["pool"] == ["f1", "f2"]
|
||||
# 持久化 roundtrip: 任务存储里的 request 保留筛选摘要, 供结果页展示
|
||||
manifest = client.store.get(body["run"]["run_id"])
|
||||
assert manifest is not None
|
||||
assert manifest["request"]["auto_screening"]["n_total"] == 61
|
||||
|
||||
|
||||
def test_auto_start_no_qualified_factors(client: Any, monkeypatch: pytest.MonkeyPatch) -> None:
|
||||
monkeypatch.setattr(
|
||||
auto_mining, "screen_all_factors",
|
||||
lambda *a, **k: _screening([]),
|
||||
)
|
||||
response = client.client.post("/api/backtest/mining/auto", json={})
|
||||
assert response.status_code == 200
|
||||
body = response.json()
|
||||
assert body["started"] is False
|
||||
assert body["reason"] == "no_qualified_factors"
|
||||
assert client.manager.start_calls == []
|
||||
|
||||
|
||||
def test_auto_start_rejects_bad_date_range(client: Any) -> None:
|
||||
response = client.client.post("/api/backtest/mining/auto", json={
|
||||
"start": "2026-09-01", "end": "2026-08-01",
|
||||
})
|
||||
assert response.status_code == 422
|
||||
@@ -14,6 +14,7 @@ from typing import ClassVar
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.indicators.pipeline import BENCHMARK_INDEX_SYMBOLS
|
||||
from app.services import quote_service as qs
|
||||
from app.services.index_const import CORE_INDEX_SYMBOLS
|
||||
|
||||
@@ -66,7 +67,7 @@ def _service_with_provider(
|
||||
monkeypatch.setattr(custom_mod, "get_provider", lambda name: provider)
|
||||
monkeypatch.setattr(
|
||||
service, "_process_full_market_records",
|
||||
lambda records, *, t0, now_ts, replace_index_cache=True: (
|
||||
lambda records, *, t0, now_ts, replace_index_cache=True, final_boundary_ms=None: (
|
||||
captured.append(records),
|
||||
index_cache_replacements.append(replace_index_cache),
|
||||
),
|
||||
@@ -84,7 +85,7 @@ def test_custom_provider_fetch_appends_index_records(monkeypatch):
|
||||
assert "600519.SH" in symbols and "000001.SH" in symbols and "399001.SZ" in symbols
|
||||
assert replacements == [True]
|
||||
# 请求清单 = 核心四只 (无指数监控规则时)
|
||||
assert provider.index_calls == [sorted(CORE_INDEX_SYMBOLS)]
|
||||
assert provider.index_calls == [sorted(set(CORE_INDEX_SYMBOLS) | BENCHMARK_INDEX_SYMBOLS)]
|
||||
|
||||
|
||||
def test_custom_provider_monitor_indices_join_fetch(monkeypatch):
|
||||
@@ -102,7 +103,7 @@ def test_custom_provider_monitor_indices_join_fetch(monkeypatch):
|
||||
service._app_state = SimpleNamespace(monitor_engine=_Engine())
|
||||
service._fetch_full_market_quotes()
|
||||
|
||||
assert provider.index_calls == [sorted(set(CORE_INDEX_SYMBOLS) | {"000300.SH"})]
|
||||
assert provider.index_calls == [sorted(set(CORE_INDEX_SYMBOLS) | BENCHMARK_INDEX_SYMBOLS | {"000300.SH"})]
|
||||
|
||||
|
||||
def test_custom_provider_without_indices_protocol_is_silent(monkeypatch):
|
||||
|
||||
@@ -42,16 +42,54 @@ def test_build_messages_contains_whitelist_fields_and_rules():
|
||||
|
||||
|
||||
def test_build_messages_only_contains_whitelisted_fields():
|
||||
from app.strategy.custom_signals import ALLOWED_FIELDS
|
||||
from app.strategy.custom_signals import allowed_fields
|
||||
|
||||
system = build_messages("x")[0]["content"]
|
||||
# 只检查「字段清单」段落（可用字段 … 运算符），排除 JSON 格式示例里的 "name"
|
||||
field_section = system.split("运算符（op）", 1)[0]
|
||||
for field in ("ma20", "rsi_14", "boll_upper"):
|
||||
assert field in field_section
|
||||
# 字段清单里出现的每个 key( 都必须在白名单内
|
||||
# 字段清单里出现的每个 key( 都必须在白名单内 (物理列 ∪ 因子)
|
||||
keys = set(re.findall(r"([a-z0-9_]+)\(", field_section))
|
||||
assert keys and keys <= ALLOWED_FIELDS
|
||||
assert keys and keys <= allowed_fields()
|
||||
|
||||
|
||||
def test_build_messages_contains_factor_fields():
|
||||
# 因子注册表字段应进入提示词 (因子·分组), 供 AI 直接构造因子条件
|
||||
from app.factors.registry import all_factors
|
||||
|
||||
system = build_messages("动量强的票")[0]["content"]
|
||||
field_section = system.split("运算符（op）", 1)[0]
|
||||
factor_ids = {s.id for s in all_factors()}
|
||||
listed = set(re.findall(r"([a-z0-9_]+)\(", field_section))
|
||||
assert factor_ids & listed, "提示词应包含至少一个因子字段"
|
||||
assert "因子·动量" in field_section
|
||||
assert "预计算的因子值" in system
|
||||
|
||||
|
||||
def test_parse_and_validate_accepts_factor_condition():
|
||||
# AI 输出以因子为条件字段: 应通过白名单校验
|
||||
raw = json.dumps({
|
||||
"name": "强动量",
|
||||
"conditions": [
|
||||
{"left": "momentum_20d", "op": ">=", "right": "0.1",
|
||||
"leftDays": 0, "rightDays": 0}
|
||||
],
|
||||
})
|
||||
result = parse_and_validate(raw)
|
||||
assert result["conditions"][0]["left"] == "momentum_20d"
|
||||
|
||||
|
||||
def test_parse_and_validate_accepts_bare_factor_rhs():
|
||||
# 右值裸写因子名: 同样自动补 field: 前缀
|
||||
raw = json.dumps({
|
||||
"name": "动量走强",
|
||||
"conditions": [
|
||||
{"left": "momentum_5d", "op": ">", "right": "momentum_20d"}
|
||||
],
|
||||
})
|
||||
result = parse_and_validate(raw)
|
||||
assert result["conditions"][0]["right"] == "field:momentum_20d"
|
||||
|
||||
|
||||
# ── parse_and_validate ──────────────────────────────────────
|
||||
|
||||
@@ -0,0 +1,92 @@
|
||||
"""自定义信号因子条件 — 字段白名单动态化 + 因子列物化链路。
|
||||
|
||||
因子接入规则 (P3): 条件可引用注册表因子 (虚拟/自定义/复合), 历史路径由
|
||||
materialize_factor_columns 复用评分物化管线补算, 与检验/评分同一条计算逻辑。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date, timedelta
|
||||
|
||||
import polars as pl
|
||||
import pytest
|
||||
|
||||
from app.strategy import custom_signals
|
||||
|
||||
|
||||
def _frame(n_days: int = 30) -> pl.DataFrame:
|
||||
"""两标的收盘价缓涨; ma20 手工预置为 0.9 倍滚动均值 (保证乖离恒为正)。"""
|
||||
rows = []
|
||||
end = date(2026, 8, 31)
|
||||
for symbol_id, base in (("A", 100.0), ("B", 50.0)):
|
||||
for i in range(n_days):
|
||||
rows.append({
|
||||
"symbol": symbol_id,
|
||||
"date": end - timedelta(days=n_days - 1 - i),
|
||||
"close": base * (1.0 + 0.001 * i),
|
||||
})
|
||||
df = pl.DataFrame(rows).sort(["symbol", "date"])
|
||||
return df.with_columns(
|
||||
(pl.col("close").rolling_mean(20).over("symbol") * 0.9).alias("ma20")
|
||||
)
|
||||
|
||||
|
||||
def test_allowed_fields_union_registry() -> None:
|
||||
allowed = custom_signals.allowed_fields()
|
||||
assert "close" in allowed # 物化白名单保留
|
||||
assert "ma20_bias" in allowed # 虚拟因子
|
||||
assert "turnover_z_60d" in allowed # 虚拟因子 (61 日预热)
|
||||
assert "momentum_20d" in allowed # 既是白名单列也是基础因子
|
||||
assert "nope_col" not in allowed
|
||||
# 静态白名单不受污染 (供 monitor_rules 等仍按物化列口径使用)
|
||||
assert "ma20_bias" not in custom_signals.ALLOWED_FIELDS
|
||||
|
||||
|
||||
def test_validate_accepts_and_rejects_factor_fields() -> None:
|
||||
ok = {
|
||||
"id": "t_bias_low", "name": "乖离超卖", "kind": "entry",
|
||||
"conditions": [{"left": "ma20_bias", "op": "<", "right": "-0.05", "leftDays": 0, "rightDays": 0}],
|
||||
}
|
||||
custom_signals.validate(ok) # 不抛错即通过
|
||||
bad = {
|
||||
"id": "t_bad", "name": "x", "kind": "entry",
|
||||
"conditions": [{"left": "not_a_field", "op": "<", "right": "1"}],
|
||||
}
|
||||
with pytest.raises(ValueError):
|
||||
custom_signals.validate(bad)
|
||||
|
||||
|
||||
def test_materialize_factor_columns_and_inject() -> None:
|
||||
sig = {
|
||||
"id": "bias_high", "name": "乖离偏高", "kind": "entry", "enabled": True,
|
||||
"conditions": [{"left": "ma20_bias", "op": ">", "right": "0", "leftDays": 0, "rightDays": 0}],
|
||||
}
|
||||
exprs = custom_signals.build_expressions([sig])
|
||||
col = custom_signals.column_name("bias_high")
|
||||
assert col in exprs
|
||||
|
||||
df = _frame()
|
||||
assert "ma20_bias" not in df.columns
|
||||
df2 = custom_signals.materialize_factor_columns(df, exprs)
|
||||
assert "ma20_bias" in df2.columns # 复用评分物化路径补算
|
||||
|
||||
# ma20 = 0.9 x 滚动均值 → 窗口内乖离恒 > 0
|
||||
warm = df2.filter(pl.col("ma20_bias").is_not_null())
|
||||
assert warm.height > 0
|
||||
assert (warm["ma20_bias"] > 0).all()
|
||||
|
||||
injected = custom_signals.inject(df2, exprs)
|
||||
assert col in injected.columns
|
||||
hit = injected.filter(pl.col("ma20_bias").is_not_null())
|
||||
assert hit[col].all() # 条件在窗口内全部成立
|
||||
|
||||
|
||||
def test_materialize_skips_unknown_columns() -> None:
|
||||
"""非注册表缺失列: 物化不处理不报错, 由 inject 缺列告警跳过。"""
|
||||
sig = {
|
||||
"id": "t_unknown", "name": "x", "kind": "entry", "enabled": True,
|
||||
"conditions": [{"left": "not_a_field", "op": "<", "right": "1", "leftDays": 0, "rightDays": 0}],
|
||||
}
|
||||
exprs = custom_signals.build_expressions([sig]) # 编译不做白名单校验 (validate 负责)
|
||||
df = _frame()
|
||||
df2 = custom_signals.materialize_factor_columns(df, exprs)
|
||||
assert df2.columns == df.columns
|
||||
@@ -11,7 +11,7 @@ from types import SimpleNamespace
|
||||
import polars as pl
|
||||
import pytest
|
||||
|
||||
from app.market_time import CN_TZ
|
||||
from app.market_time import CN_TZ, cn_today
|
||||
from app.services.data_integrity import (
|
||||
AUTO_REPAIR_MAX_LAG_DAYS,
|
||||
IntegrityIssue,
|
||||
@@ -189,19 +189,61 @@ def test_realtime_daily_builder_drops_halted_rows_before_zero_fill():
|
||||
"symbol": "600001.SH", "last_price": 10.0,
|
||||
"open": 9.9, "high": 10.1, "low": 9.8,
|
||||
"volume": 1000.0, "amount": 10000.0,
|
||||
"timestamp": _ts_ms(TODAY, time(10, 0)),
|
||||
"timestamp": _ts_ms(cn_today(), time(10, 0)),
|
||||
},
|
||||
{
|
||||
"symbol": "600002.SH", "last_price": 20.0,
|
||||
"open": 0.0, "high": 0.0, "low": 0.0,
|
||||
"volume": 0.0, "amount": 0.0,
|
||||
"timestamp": _ts_ms(TODAY, time(9, 15)),
|
||||
"timestamp": _ts_ms(cn_today(), time(9, 15)),
|
||||
},
|
||||
])
|
||||
|
||||
assert result["symbol"].to_list() == ["600001.SH"]
|
||||
|
||||
|
||||
def test_realtime_daily_builder_drops_stale_snapshot_rows():
|
||||
"""回归: 停牌股快照停留旧日, 不得复制成当日假蜡烛 (301266.SZ 2026-09-04)。
|
||||
|
||||
实时源对停牌标的返回停牌前最后一份快照 — OHLCV 全为旧日真实值, 仅
|
||||
timestamp 停在旧日。这种记录不属于当日, 必须在落盘前按 quote_ts 过滤。
|
||||
"""
|
||||
from app.services.quote_service import QuoteService
|
||||
|
||||
halted_since = cn_today() - timedelta(days=7)
|
||||
result = QuoteService._build_daily([
|
||||
{
|
||||
"symbol": "600001.SH", "last_price": 10.0,
|
||||
"open": 9.9, "high": 10.1, "low": 9.8,
|
||||
"volume": 1000.0, "amount": 10000.0,
|
||||
"timestamp": _ts_ms(cn_today(), time(15, 0)),
|
||||
},
|
||||
{
|
||||
"symbol": "301266.SZ", "last_price": 24.97,
|
||||
"open": 23.10, "high": 24.99, "low": 23.01,
|
||||
"volume": 65038.0, "amount": 157796300.0,
|
||||
"timestamp": _ts_ms(halted_since, time(15, 30)),
|
||||
},
|
||||
])
|
||||
|
||||
assert result["symbol"].to_list() == ["600001.SH"]
|
||||
|
||||
|
||||
def test_realtime_daily_builder_keeps_rows_without_timestamp():
|
||||
"""无时间戳的源无法判断快照新旧, 维持原行为保留 (不因缺列误删)。"""
|
||||
from app.services.quote_service import QuoteService
|
||||
|
||||
result = QuoteService._build_daily([
|
||||
{
|
||||
"symbol": "600003.SH", "last_price": 8.0,
|
||||
"open": 7.9, "high": 8.1, "low": 7.8,
|
||||
"volume": 500.0, "amount": 4000.0,
|
||||
},
|
||||
])
|
||||
|
||||
assert result["symbol"].to_list() == ["600003.SH"]
|
||||
|
||||
|
||||
def test_halt_filter_drops_legacy_zero_volume_row_after_ohlc_fill():
|
||||
from app.indicators.pipeline import filter_halt_days
|
||||
|
||||
|
||||
@@ -0,0 +1,83 @@
|
||||
"""收盘价过期分区回归: 盘后管道必须重算"行数完整但价格停留在竞价前"的 enriched 分区。
|
||||
|
||||
实时 flush 可能在收盘集合竞价结果发布前写入当日分区 (实测 2026-09-04: TickFlow
|
||||
实时端点收盘后仍返回旧价, 3392/5554 只股票 enriched 收盘价与官方日线不符),
|
||||
分区行数与 daily 相同, #223 的行数校验识别不到, 增量路径不会重算, 当日偏离值/
|
||||
动量等全部 enriched 消费方都会用旧价 (海鸥住工 10 日偏离 98.77% vs 官方 99.61%)。
|
||||
_prune_stale_price_partitions 按官方日线做 raw_close vs close 值级比对, 不一致即
|
||||
删分区, 让增量重算按官方日线全市场重建。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.jobs.daily_pipeline import _prune_stale_price_partitions
|
||||
|
||||
|
||||
def _write_partition(base: Path, day: str, symbols: list[str], closes: list[float], col: str) -> None:
|
||||
part = base / f"date={day}"
|
||||
part.mkdir(parents=True, exist_ok=True)
|
||||
pl.DataFrame({"symbol": symbols, col: closes}).write_parquet(part / "part.parquet")
|
||||
|
||||
|
||||
def test_stale_close_partition_is_pruned(tmp_path) -> None:
|
||||
daily = tmp_path / "kline_daily"
|
||||
enriched = tmp_path / "kline_daily_enriched"
|
||||
syms = ["002084.SZ", "600519.SH", "000001.SZ"]
|
||||
# 2026-09-04: 官方日线收盘 7.10, 实时写入的 enriched 停留在竞价前 7.07;
|
||||
# 行数两侧一致 (覆盖全), #223 行数校验识别不到 (issue 实测形态)
|
||||
_write_partition(daily, "2026-09-04", syms, [7.10, 1330.10, 11.89], "close")
|
||||
_write_partition(enriched, "2026-09-04", syms, [7.07, 1330.10, 11.89], "raw_close")
|
||||
|
||||
pruned = _prune_stale_price_partitions(daily, enriched)
|
||||
|
||||
assert pruned == ["2026-09-04"]
|
||||
assert not (enriched / "date=2026-09-04").exists()
|
||||
|
||||
|
||||
def test_matching_close_partitions_untouched(tmp_path) -> None:
|
||||
daily = tmp_path / "kline_daily"
|
||||
enriched = tmp_path / "kline_daily_enriched"
|
||||
syms = ["002084.SZ", "600519.SH"]
|
||||
_write_partition(daily, "2026-09-04", syms, [7.10, 1330.10], "close")
|
||||
_write_partition(enriched, "2026-09-04", syms, [7.10, 1330.10], "raw_close")
|
||||
|
||||
assert _prune_stale_price_partitions(daily, enriched) == []
|
||||
assert (enriched / "date=2026-09-04" / "part.parquet").exists()
|
||||
|
||||
|
||||
def test_multiple_stale_dates_all_pruned(tmp_path) -> None:
|
||||
# 管道连续数日未触发重建时, 最近多个交易日的过期分区一并修复
|
||||
daily = tmp_path / "kline_daily"
|
||||
enriched = tmp_path / "kline_daily_enriched"
|
||||
for day, stale in [("2026-09-04", 7.07), ("2026-09-03", 6.88), ("2026-09-02", 6.80)]:
|
||||
_write_partition(daily, day, ["002084.SZ"], [stale + 0.03], "close")
|
||||
_write_partition(enriched, day, ["002084.SZ"], [stale], "raw_close")
|
||||
|
||||
pruned = _prune_stale_price_partitions(daily, enriched)
|
||||
|
||||
assert sorted(pruned) == ["2026-09-02", "2026-09-03", "2026-09-04"]
|
||||
|
||||
|
||||
def test_missing_raw_close_column_left_alone(tmp_path) -> None:
|
||||
# 旧 schema 无 raw_close 列 → 读列失败, 交给既有完整性检查, 不误删
|
||||
daily = tmp_path / "kline_daily"
|
||||
enriched = tmp_path / "kline_daily_enriched"
|
||||
_write_partition(daily, "2026-09-04", ["002084.SZ"], [7.10], "close")
|
||||
part = enriched / "date=2026-09-04"
|
||||
part.mkdir(parents=True)
|
||||
pl.DataFrame({"symbol": ["002084.SZ"], "close": [7.07]}).write_parquet(part / "part.parquet")
|
||||
|
||||
assert _prune_stale_price_partitions(daily, enriched) == []
|
||||
assert (part / "part.parquet").exists()
|
||||
|
||||
|
||||
def test_daily_partition_missing_left_alone(tmp_path) -> None:
|
||||
# 官方日线尚未同步的日期不比对 (留给当日正常流程)
|
||||
enriched = tmp_path / "kline_daily_enriched"
|
||||
_write_partition(enriched, "2026-09-04", ["002084.SZ"], [7.07], "raw_close")
|
||||
|
||||
assert _prune_stale_price_partitions(tmp_path / "kline_daily", enriched) == []
|
||||
assert (enriched / "date=2026-09-04" / "part.parquet").exists()
|
||||
@@ -0,0 +1,309 @@
|
||||
"""因子 API (validate / trial) 契约测试 (P2)。"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date, timedelta
|
||||
|
||||
import polars as pl
|
||||
import pytest
|
||||
from fastapi import FastAPI
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from app.api.factors import router
|
||||
|
||||
|
||||
@pytest.fixture()
|
||||
def cleanup_registry():
|
||||
"""测试注册的自定义因子在用例后注销, 不污染全局注册表 (快照测试依赖 77 基线)。"""
|
||||
created: set[str] = set()
|
||||
yield created
|
||||
from app.factors.registry import unregister_factor
|
||||
|
||||
for fid in created:
|
||||
unregister_factor(fid)
|
||||
|
||||
|
||||
class _FakeEngine:
|
||||
"""合成面板: 3 只股票日收益固定 1%/2%/3%, 任何按价格排序的因子 IC 恒为 1。"""
|
||||
|
||||
def __init__(self, n_days: int = 40) -> None:
|
||||
rows = []
|
||||
end = date.today()
|
||||
for index in range(n_days):
|
||||
day = end - timedelta(days=n_days - 1 - index)
|
||||
for symbol_id, daily_return in (("A", 0.01), ("B", 0.02), ("C", 0.03)):
|
||||
# 正基数且增速同序: C 永远最高价且回报最高 → 按价格排序的因子 IC 恒为 1
|
||||
rows.append({
|
||||
"symbol": symbol_id,
|
||||
"date": day,
|
||||
"close": (1.0 + daily_return) ** index * 10.0 * (ord(symbol_id) - ord("A") + 1),
|
||||
})
|
||||
self.panel = pl.DataFrame(rows).sort(["symbol", "date"])
|
||||
|
||||
def load_panel(self, symbols, start, end, *, columns=None, asset_type="stock", **_kwargs):
|
||||
frame = self.panel
|
||||
if columns is not None:
|
||||
for column in columns:
|
||||
if column not in frame.columns:
|
||||
frame = frame.with_columns(pl.lit(None).cast(pl.Float64).alias(column))
|
||||
frame = frame.select(columns)
|
||||
return frame.filter((pl.col("date") >= start) & (pl.col("date") <= end))
|
||||
|
||||
|
||||
def _client(with_engine: bool = False) -> TestClient:
|
||||
app = FastAPI()
|
||||
app.include_router(router)
|
||||
if with_engine:
|
||||
app.state.backtest_engine = _FakeEngine()
|
||||
return TestClient(app)
|
||||
|
||||
|
||||
def test_validate_ok_formula() -> None:
|
||||
response = _client().post("/api/factors/validate", json={"formula": "rank(-ts_sum(change_pct, 5))"})
|
||||
assert response.status_code == 200
|
||||
payload = response.json()
|
||||
assert payload["ok"] is True
|
||||
assert payload["errors"] == []
|
||||
assert payload["dependencies"] == ["change_pct"]
|
||||
assert payload["warmup_bars"] == 6
|
||||
assert payload["cross_sectional"] is True
|
||||
|
||||
|
||||
def test_validate_future_function_rejected() -> None:
|
||||
response = _client().post("/api/factors/validate", json={"formula": "rank(ts_delta(close, -5))"})
|
||||
assert response.status_code == 200
|
||||
payload = response.json()
|
||||
assert payload["ok"] is False
|
||||
assert payload["errors"][0]["code"] == "E005"
|
||||
assert "position" in payload["errors"][0]
|
||||
|
||||
|
||||
def test_trial_golden_ic() -> None:
|
||||
response = _client(with_engine=True).post(
|
||||
"/api/factors/trial", json={"formula": "close", "days": 30},
|
||||
)
|
||||
assert response.status_code == 200
|
||||
payload = response.json()
|
||||
assert payload["ok"] is True
|
||||
assert payload["n_dates"] == 30 # 40 日面板, 首日无前收 → 39 个可算截面, 取最近 30
|
||||
assert payload["ic_mean"] == pytest.approx(1.0, abs=1e-9)
|
||||
assert payload["ic_win_rate"] == pytest.approx(1.0, abs=1e-9)
|
||||
# 恒定 IC 序列 std=0, IR 无定义 → None (除零保护)
|
||||
assert payload["ic_std"] in (None, 0.0)
|
||||
assert payload["ir"] is None
|
||||
assert len(payload["ic_series"]) == 30
|
||||
|
||||
|
||||
def test_trial_compile_failure_400() -> None:
|
||||
response = _client(with_engine=True).post(
|
||||
"/api/factors/trial", json={"formula": "nope_col + 1", "days": 30},
|
||||
)
|
||||
assert response.status_code == 400
|
||||
body = response.json()["detail"]
|
||||
assert body["errors"][0]["code"] == "E001"
|
||||
|
||||
|
||||
def test_trial_computes_virtual_factor_via_shared_path() -> None:
|
||||
# 引用虚拟因子 ma20_bias: 试算端点复用 _compute_missing_factors 补算路径
|
||||
# (compute_indicators 算 ma20 + materialize 物化 bias), 3 列粗面板即可出结果
|
||||
response = _client(with_engine=True).post(
|
||||
"/api/factors/trial", json={"formula": "ma20_bias", "days": 20},
|
||||
)
|
||||
assert response.status_code == 200
|
||||
payload = response.json()
|
||||
assert payload["ok"] is True
|
||||
assert payload["n_dates"] == 20
|
||||
assert payload["ic_mean"] is not None
|
||||
|
||||
|
||||
def test_group_and_status_update_after_registry_load(tmp_path, cleanup_registry) -> None:
|
||||
"""改分组/状态在因子已注册 (启动加载后) 的真实路径下可用。
|
||||
|
||||
回归保护: 同版本直接 register 会被注册表拒绝 ("版本未提升"),
|
||||
端点必须先注销再按新元数据注册。
|
||||
"""
|
||||
from pathlib import Path
|
||||
from types import SimpleNamespace
|
||||
|
||||
from app.factors.registry import get_factor
|
||||
|
||||
app = FastAPI()
|
||||
app.include_router(router)
|
||||
app.state.backtest_engine = _FakeEngine()
|
||||
app.state.repo = SimpleNamespace(store=SimpleNamespace(data_dir=Path(tmp_path)))
|
||||
client = TestClient(app)
|
||||
|
||||
created = client.post("/api/factors/custom", json={
|
||||
"id": "uf_group_test", "label": "分组测试", "formula": "rank(-ts_sum(change_pct, 5))",
|
||||
})
|
||||
assert created.status_code == 200
|
||||
factor_id = created.json()["id"]
|
||||
cleanup_registry.add(factor_id)
|
||||
|
||||
from app.factors import store
|
||||
|
||||
store.load_into_registry(Path(tmp_path)) # 模拟重启后的注册状态
|
||||
registered = get_factor(factor_id)
|
||||
assert registered is not None and registered.group == "自定义"
|
||||
|
||||
renamed = client.post(f"/api/factors/custom/{factor_id}/group", json={"group": "我的动量组"})
|
||||
assert renamed.status_code == 200
|
||||
assert renamed.json()["group"] == "我的动量组"
|
||||
refreshed = get_factor(factor_id)
|
||||
assert refreshed is not None and refreshed.group == "我的动量组" # 注册表同步
|
||||
on_disk = next(d for d in store.load_all(Path(tmp_path)) if d["id"] == factor_id)
|
||||
assert on_disk["group"] == "我的动量组" # 磁盘持久化
|
||||
|
||||
activated = client.post(f"/api/factors/custom/{factor_id}/status", json={"status": "active"})
|
||||
assert activated.status_code == 200 # 修复前: 400 "版本未提升"
|
||||
assert get_factor(factor_id).stability == "stable"
|
||||
|
||||
bad = client.post(f"/api/factors/custom/{factor_id}/group", json={"group": " "})
|
||||
assert bad.status_code == 400 # 空白分组名 fail-closed
|
||||
|
||||
|
||||
def test_update_custom_factor_bumps_version(tmp_path, cleanup_registry) -> None:
|
||||
"""编辑已有自定义因子: 版本提升注册 + 公式变化回 draft + 试算门禁。"""
|
||||
from pathlib import Path
|
||||
from types import SimpleNamespace
|
||||
|
||||
from app.factors import store
|
||||
from app.factors.registry import get_factor
|
||||
|
||||
cleanup_registry.add("uf_edit_test")
|
||||
app = FastAPI()
|
||||
app.include_router(router)
|
||||
app.state.backtest_engine = _FakeEngine()
|
||||
app.state.repo = SimpleNamespace(store=SimpleNamespace(data_dir=Path(tmp_path)))
|
||||
client = TestClient(app)
|
||||
data_dir = Path(tmp_path)
|
||||
|
||||
created = client.post("/api/factors/custom", json={
|
||||
"id": "uf_edit_test", "label": "编辑测试", "formula": "rank(-ts_sum(change_pct, 5))",
|
||||
})
|
||||
assert created.status_code == 200
|
||||
assert created.json()["version"] == 1
|
||||
|
||||
# 激活后再编辑: 公式变化 → 新版本 + 回 draft (生命周期语义)
|
||||
activated = client.post("/api/factors/custom/uf_edit_test/status", json={"status": "active"})
|
||||
assert activated.status_code == 200
|
||||
|
||||
updated = client.post("/api/factors/custom/uf_edit_test/update", json={
|
||||
"label": "编辑测试v2", "group": "新分组", "formula": "rank(-ts_sum(change_pct, 10))",
|
||||
"description": "窗口从 5 改 10", "direction": "low",
|
||||
})
|
||||
assert updated.status_code == 200
|
||||
body = updated.json()
|
||||
assert body["version"] == 2 and body["status"] == "draft"
|
||||
|
||||
spec = get_factor("uf_edit_test")
|
||||
assert spec is not None
|
||||
assert spec.version == 2 and spec.group == "新分组" and spec.label == "编辑测试v2"
|
||||
assert "change_pct" in spec.dependencies
|
||||
on_disk = next(d for d in store.load_all(data_dir) if d["id"] == "uf_edit_test")
|
||||
assert on_disk["version"] == 2 and on_disk["status"] == "draft"
|
||||
|
||||
# 仅改元数据 (公式不变): 版本仍提升, 状态保留 (不回 draft)
|
||||
client.post("/api/factors/custom/uf_edit_test/status", json={"status": "active"})
|
||||
meta = client.post("/api/factors/custom/uf_edit_test/update", json={
|
||||
"label": "仅改名字", "group": "新分组", "formula": "rank(-ts_sum(change_pct, 10))",
|
||||
})
|
||||
assert meta.status_code == 200
|
||||
assert meta.json() == {"ok": True, "id": "uf_edit_test", "version": 3, "status": "active"}
|
||||
|
||||
missing = client.post("/api/factors/custom/uf_nope/update", json={
|
||||
"label": "x", "formula": "close",
|
||||
})
|
||||
assert missing.status_code == 404
|
||||
|
||||
|
||||
def test_trial_response_includes_newey_west_t() -> None:
|
||||
response = _client(with_engine=True).post(
|
||||
"/api/factors/trial", json={"formula": "close", "days": 30},
|
||||
)
|
||||
assert response.status_code == 200
|
||||
payload = response.json()
|
||||
assert payload["n_dates"] == 30
|
||||
assert "t_newey_west" in payload # 恒定 IC 序列下可为 None, 但字段必须存在
|
||||
|
||||
|
||||
def test_delete_custom_and_composite_factor(tmp_path, cleanup_registry) -> None:
|
||||
"""删除契约: 复合因子可直接删; 成员被复合引用时 409 列引用方, force 才放行。"""
|
||||
from pathlib import Path
|
||||
from types import SimpleNamespace
|
||||
|
||||
from app.factors import store
|
||||
from app.factors.registry import get_factor
|
||||
|
||||
app = FastAPI()
|
||||
app.include_router(router)
|
||||
app.state.backtest_engine = _FakeEngine() # 创建门禁需试算
|
||||
app.state.repo = SimpleNamespace(store=SimpleNamespace(data_dir=Path(tmp_path)))
|
||||
client = TestClient(app)
|
||||
data_dir = Path(tmp_path)
|
||||
|
||||
client.post("/api/factors/custom", json={
|
||||
"id": "uf_del_member", "label": "被引用成员", "formula": "rank(-ts_sum(change_pct, 5))",
|
||||
})
|
||||
cleanup_registry.add("uf_del_member")
|
||||
created = client.post("/api/factors/composite", json={
|
||||
"id": "cf_del_test", "label": "删除测试组合", "members": {"uf_del_member": 1.0, "momentum_10d": -0.5},
|
||||
})
|
||||
assert created.status_code == 200
|
||||
cleanup_registry.add("cf_del_test")
|
||||
|
||||
# 404: 不存在的因子
|
||||
assert client.delete("/api/factors/custom/cf_nope").status_code == 404
|
||||
|
||||
# 409: 成员被复合因子引用 → fail-closed, 返回引用方列表
|
||||
blocked = client.delete("/api/factors/custom/uf_del_member")
|
||||
assert blocked.status_code == 409
|
||||
detail = blocked.json()["detail"]
|
||||
assert "cf_del_test" in str(detail["references"])
|
||||
assert get_factor("uf_del_member") is not None # 引用未解除, 因子仍在
|
||||
|
||||
# 复合因子本身可直接删除 (无人引用它)
|
||||
removed = client.delete("/api/factors/custom/cf_del_test")
|
||||
assert removed.status_code == 200
|
||||
assert removed.json() == {"ok": True, "id": "cf_del_test", "removed_references": []}
|
||||
assert get_factor("cf_del_test") is None
|
||||
assert all(d["id"] != "cf_del_test" for d in store.load_all(data_dir)) # 磁盘已删
|
||||
|
||||
# 引用解除后成员可正常删除
|
||||
freed = client.delete("/api/factors/custom/uf_del_member")
|
||||
assert freed.status_code == 200
|
||||
assert get_factor("uf_del_member") is None
|
||||
|
||||
|
||||
def test_delete_with_strategy_reference_requires_force(tmp_path, cleanup_registry) -> None:
|
||||
"""策略文件引用同样拦截: 409 列出 strategies/*.json, force=true 放行。"""
|
||||
import json
|
||||
from pathlib import Path
|
||||
from types import SimpleNamespace
|
||||
|
||||
from app.factors.registry import get_factor
|
||||
|
||||
app = FastAPI()
|
||||
app.include_router(router)
|
||||
app.state.backtest_engine = _FakeEngine() # 创建门禁需试算
|
||||
app.state.repo = SimpleNamespace(store=SimpleNamespace(data_dir=Path(tmp_path)))
|
||||
client = TestClient(app)
|
||||
data_dir = Path(tmp_path)
|
||||
|
||||
client.post("/api/factors/custom", json={
|
||||
"id": "uf_strat_ref", "label": "策略引用", "formula": "rank(-ts_sum(change_pct, 5))",
|
||||
})
|
||||
cleanup_registry.add("uf_strat_ref")
|
||||
strategies_dir = data_dir / "strategies"
|
||||
strategies_dir.mkdir()
|
||||
(strategies_dir / "my_strategy.json").write_text(
|
||||
json.dumps({"name": "my_strategy", "factors": {"uf_strat_ref": 1.0}}), encoding="utf-8",
|
||||
)
|
||||
|
||||
blocked = client.delete("/api/factors/custom/uf_strat_ref")
|
||||
assert blocked.status_code == 409
|
||||
assert blocked.json()["detail"]["references"] == ["strategies/my_strategy.json"]
|
||||
|
||||
forced = client.delete("/api/factors/custom/uf_strat_ref?force=true")
|
||||
assert forced.status_code == 200
|
||||
assert forced.json()["removed_references"] == ["strategies/my_strategy.json"]
|
||||
assert get_factor("uf_strat_ref") is None
|
||||
@@ -0,0 +1,80 @@
|
||||
"""回测因子归因 (v1) — _factor_attribution_summary 单元测试。
|
||||
|
||||
覆盖:
|
||||
- 盈利/亏损单因子均值、样本数计算
|
||||
- snapshot 日期列 date/str 两种 dtype 均可关联
|
||||
- entry_signal_date 缺失时回退 entry_date
|
||||
- 无可关联行 / 空成交 / 无因子列 → None (fail-open)
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date
|
||||
from types import SimpleNamespace
|
||||
|
||||
import polars as pl
|
||||
|
||||
from app.backtest.strategy import _factor_attribution_summary
|
||||
|
||||
|
||||
def _trade(symbol: str, day: str, pnl: float):
|
||||
return SimpleNamespace(
|
||||
symbol=symbol,
|
||||
entry_signal_date=day,
|
||||
entry_date=day,
|
||||
pnl_pct=pnl,
|
||||
)
|
||||
|
||||
|
||||
def _snapshot(dates_as: str = "str") -> pl.DataFrame:
|
||||
frame = pl.DataFrame({
|
||||
"symbol": ["000001", "000002", "000003", "000004"],
|
||||
"date": ["2026-01-05"] * 4,
|
||||
"momentum_20d": [0.10, -0.05, 0.20, 0.00],
|
||||
"turnover_rate": [5.0, 8.0, 6.0, 7.0],
|
||||
})
|
||||
if dates_as == "date":
|
||||
frame = frame.with_columns(pl.col("date").str.to_date())
|
||||
return frame
|
||||
|
||||
|
||||
def test_summary_win_lose_means():
|
||||
trades = [
|
||||
_trade("000001", "2026-01-05", 0.10), # 盈利: mom 0.10, to 5
|
||||
_trade("000003", "2026-01-05", 0.05), # 盈利: mom 0.20, to 6
|
||||
_trade("000002", "2026-01-05", -0.03), # 亏损: mom -0.05, to 8
|
||||
_trade("000004", "2026-01-05", -0.08), # 亏损: mom 0.00, to 7
|
||||
]
|
||||
result = _factor_attribution_summary(_snapshot(), trades)
|
||||
assert result is not None
|
||||
assert result["n_win"] == 2 and result["n_lose"] == 2
|
||||
by_factor = {f["factor"]: f for f in result["factors"]}
|
||||
assert by_factor["momentum_20d"]["win_mean"] == round((0.10 + 0.20) / 2, 6)
|
||||
assert by_factor["momentum_20d"]["lose_mean"] == round((-0.05 + 0.00) / 2, 6)
|
||||
assert by_factor["turnover_rate"]["win_mean"] == 5.5
|
||||
assert by_factor["turnover_rate"]["lose_n"] == 2
|
||||
|
||||
|
||||
def test_summary_accepts_date_dtype_snapshot():
|
||||
trades = [_trade("000001", "2026-01-05", 0.1), _trade("000002", "2026-01-05", -0.1)]
|
||||
result = _factor_attribution_summary(_snapshot(dates_as="date"), trades)
|
||||
assert result is not None
|
||||
assert result["n_win"] == 1
|
||||
|
||||
|
||||
def test_summary_falls_back_to_entry_date():
|
||||
trade = SimpleNamespace(symbol="000001", entry_signal_date=None,
|
||||
entry_date=date(2026, 1, 5), pnl_pct=0.2)
|
||||
result = _factor_attribution_summary(_snapshot(), [trade])
|
||||
assert result is not None
|
||||
assert result["factors"][0]["win_n"] == 1
|
||||
|
||||
|
||||
def test_summary_returns_none_when_no_overlap():
|
||||
trades = [_trade("600000", "2026-02-10", 0.1)] # 不在快照里
|
||||
assert _factor_attribution_summary(_snapshot(), trades) is None
|
||||
|
||||
|
||||
def test_summary_returns_none_on_empty_inputs():
|
||||
assert _factor_attribution_summary(_snapshot(), []) is None
|
||||
no_factor = pl.DataFrame({"symbol": ["000001"], "date": ["2026-01-05"]})
|
||||
assert _factor_attribution_summary(no_factor, [_trade("000001", "2026-01-05", 0.1)]) is None
|
||||
@@ -0,0 +1,209 @@
|
||||
"""因子公式 DSL 编译器测试 (P2)。
|
||||
|
||||
覆盖: 全部编译期错误码 E001-E008/E010/E011/E014/E016、窗口纪律 (只向后看)、
|
||||
数值正确性 (与手算基准对拍)、依赖/预热推导、运行期 fail-closed。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date
|
||||
|
||||
import polars as pl
|
||||
import pytest
|
||||
|
||||
from app.factors.dsl import BASE_COLUMNS, FACTOR_COLUMN, compile_formula
|
||||
|
||||
|
||||
def _panel() -> pl.DataFrame:
|
||||
# 两个 symbol x 6 日, 便于验证 over("symbol") 不串组
|
||||
rows = []
|
||||
for symbol, closes in (("A", [10.0, 11.0, 12.0, 13.0, 14.0, 15.0]),
|
||||
("B", [100.0, 90.0, 80.0, 70.0, 60.0, 50.0])):
|
||||
for index, close in enumerate(closes):
|
||||
rows.append({
|
||||
"symbol": symbol,
|
||||
"date": date(2026, 1, index + 1),
|
||||
"close": close,
|
||||
"volume": 1000.0 + index * 100,
|
||||
"amount": (1000.0 + index * 100) * close,
|
||||
})
|
||||
return pl.DataFrame(rows).sort(["symbol", "date"])
|
||||
|
||||
|
||||
def _eval(formula: str, panel: pl.DataFrame | None = None) -> pl.Series:
|
||||
compiled = compile_formula(formula)
|
||||
assert compiled.ok, [error.to_dict() for error in compiled.errors]
|
||||
assert compiled.frame_transform is not None
|
||||
base = panel if panel is not None else _panel()
|
||||
frame = compiled.frame_transform(base)
|
||||
assert frame is not None
|
||||
return frame[FACTOR_COLUMN]
|
||||
|
||||
|
||||
# ------------------------------------------------------------- 错误码覆盖
|
||||
|
||||
@pytest.mark.parametrize("formula, code", [
|
||||
("clos + 1", "E001"), # 未知标识符
|
||||
("foo(close)", "E002"), # 未知函数
|
||||
("ts_mean(close)", "E003"), # 缺窗口参数
|
||||
("ts_mean(close, close)", "E003"), # 窗口参数必须是常量
|
||||
("ts_mean(close, 1)", "E004"), # 窗口 < 2
|
||||
("ts_mean(close, 600)", "E004"), # 窗口 > 512
|
||||
("ts_quantile(close, 5, 1.5)", "E004"), # q 不在 (0,1)
|
||||
("ts_delay(close, -5)", "E005"), # 负 shift = 未来函数
|
||||
("ts_delta(close, -1)", "E005"),
|
||||
("power(close, 5)", "E010"), # 指数越界
|
||||
("winsorize(close, 9)", "E011"), # k 越界
|
||||
("close / 0", "E008"), # 静态除零
|
||||
("1 + 2", "E016"), # 常量表达式
|
||||
("close +", "E014"), # 语法错误
|
||||
("(close", "E014"),
|
||||
("close $ 1", "E014"),
|
||||
("", "E014"),
|
||||
])
|
||||
def test_error_codes(formula: str, code: str) -> None:
|
||||
compiled = compile_formula(formula)
|
||||
assert not compiled.ok
|
||||
assert any(error.code == code for error in compiled.errors), [e.code for e in compiled.errors]
|
||||
|
||||
|
||||
def test_depth_and_token_limits() -> None:
|
||||
# AST 深度: 嵌套 13 层二元运算 (每层 bin 算 1) 超限
|
||||
deep = "close"
|
||||
for _ in range(13):
|
||||
deep = f"({deep} + 1)"
|
||||
compiled = compile_formula(deep)
|
||||
assert not compiled.ok
|
||||
assert any(error.code == "E006" for error in compiled.errors)
|
||||
|
||||
wide = " + ".join(["close"] * 120) # 二元链深度为 2, 但 token 超 200
|
||||
compiled = compile_formula(wide)
|
||||
assert any(error.code == "E007" for error in compiled.errors)
|
||||
|
||||
|
||||
def test_error_payload_shape() -> None:
|
||||
compiled = compile_formula("rank(ts_delta(close, -5))")
|
||||
assert not compiled.ok
|
||||
payload = compiled.errors[0].to_dict()
|
||||
assert payload["code"] == "E005"
|
||||
assert payload["message"]
|
||||
assert "position" in payload and "offset" in payload["position"]
|
||||
|
||||
|
||||
# ------------------------------------------------------------- 依赖与预热
|
||||
|
||||
def test_dependencies_and_warmup() -> None:
|
||||
compiled = compile_formula("rank(-ts_sum(change_pct, 5))")
|
||||
assert compiled.ok
|
||||
# change_pct 是 base 因子, 依赖为其自身列
|
||||
assert compiled.dependencies == frozenset({"change_pct"})
|
||||
assert compiled.referenced_factors == frozenset({"change_pct"})
|
||||
assert compiled.warmup_bars == 6 # ts 窗口 5 + 1
|
||||
assert compiled.cross_sectional
|
||||
|
||||
compiled = compile_formula("close + ma20_bias")
|
||||
assert compiled.ok
|
||||
assert compiled.dependencies == frozenset({"close", "ma20"})
|
||||
assert compiled.referenced_factors == frozenset({"ma20_bias"})
|
||||
|
||||
compiled = compile_formula("turnover_z_60d * 2")
|
||||
assert compiled.ok
|
||||
assert compiled.warmup_bars == 61 # 引用因子 warmup 传递
|
||||
|
||||
|
||||
def test_base_columns_contract() -> None:
|
||||
assert "close" in BASE_COLUMNS
|
||||
assert "clos" not in BASE_COLUMNS
|
||||
|
||||
|
||||
# ------------------------------------------------------------- 数值正确性
|
||||
|
||||
def test_ts_delay_backward_only() -> None:
|
||||
values = _eval("ts_delay(close, 2)")
|
||||
# A 组: [null, null, 10, 11, 12, 13]; B 组: [null, null, 100, 90, 80, 70]
|
||||
assert values[:6].to_list() == [None, None, 10.0, 11.0, 12.0, 13.0]
|
||||
assert values[6:].to_list() == [None, None, 100.0, 90.0, 80.0, 70.0]
|
||||
|
||||
|
||||
def test_ts_mean_no_cross_symbol_leak() -> None:
|
||||
values = _eval("ts_mean(close, 2)")
|
||||
# A 组 2 日窗: [null, 10.5, 11.5, 12.5, 13.5, 14.5]
|
||||
assert values[:6].to_list() == [None, 10.5, 11.5, 12.5, 13.5, 14.5]
|
||||
# B 组边界: 第一行是 null (窗口不满) 而不是拿到 A 组尾部; 第二行 95.0
|
||||
assert values[6] is None
|
||||
assert values[7] == 95.0
|
||||
|
||||
|
||||
def test_rank_cross_sectional() -> None:
|
||||
values = _eval("rank(close)")
|
||||
frame = _panel().with_columns(pl.Series("_f", values))
|
||||
per_date = frame.filter(pl.col("date") == date(2026, 1, 1))
|
||||
# 第一日 A=10, B=100: rank(A) < rank(B), 且都 ∈ (0,1]
|
||||
ranks = dict(zip(per_date["symbol"].to_list(), per_date["_f"].to_list(), strict=True))
|
||||
assert 0 < ranks["A"] < ranks["B"] <= 1.0
|
||||
|
||||
|
||||
def test_cross_of_timeseries_two_phase() -> None:
|
||||
# Polars 嵌套窗口会静默全 null; 编译器必须走两阶段 (临时列 + 截面)
|
||||
values = _eval("rank(ts_sum(close, 2))")
|
||||
assert sum(1 for value in values.to_list() if value is not None) > 0
|
||||
# B 组滚动和数值恒大于 A 组 (百元 vs 十元价位) → 每个 (非首行) 日期 rank(B) > rank(A)
|
||||
frame = _panel().with_columns(pl.Series("_f", values))
|
||||
for day in (date(2026, 1, 3), date(2026, 1, 6)):
|
||||
day_frame = frame.filter(pl.col("date") == day)
|
||||
ranks = dict(zip(day_frame["symbol"].to_list(), day_frame["_f"].to_list(), strict=True))
|
||||
assert ranks["B"] > ranks["A"]
|
||||
|
||||
|
||||
def test_cross_in_timeseries_rejected() -> None:
|
||||
compiled = compile_formula("ts_mean(rank(close), 5)")
|
||||
assert not compiled.ok
|
||||
assert any(error.code == "E009" for error in compiled.errors)
|
||||
|
||||
|
||||
def test_safe_division_yields_null() -> None:
|
||||
# 分母为动态表达式的恒 0: 静态折叠不报 E008, 运行期安全除产出 null
|
||||
values = _eval("close / (close - close)")
|
||||
assert all(value is None for value in values.to_list())
|
||||
|
||||
|
||||
def test_if_else_and_comparison() -> None:
|
||||
values = _eval("if_else(close > 12, 1, 0)")
|
||||
assert values[:6].to_list() == [0.0, 0.0, 0.0, 1.0, 1.0, 1.0]
|
||||
|
||||
|
||||
def test_arithmetic_precedence() -> None:
|
||||
values = _eval("close + 2 * 3")
|
||||
assert values[0] == 16.0 # 10 + 6, 而不是 (10+2)*3
|
||||
|
||||
|
||||
def test_decay_linear_weights_recent() -> None:
|
||||
values = _eval("decay_linear(close, 3)")
|
||||
# A 组第 3 行: (3*12 + 2*11 + 1*10) / 6 = 68/6
|
||||
assert values[2] == pytest.approx((3 * 12 + 2 * 11 + 1 * 10) / 6)
|
||||
|
||||
|
||||
def test_log_domain_guard() -> None:
|
||||
import math
|
||||
|
||||
panel = _panel().with_columns((pl.col("close") - 15.0).alias("neg"))
|
||||
values = _eval("log(close - 15)", panel)
|
||||
# A 组全为负 → null; B 组 100-15=85 → log 正常
|
||||
assert all(value is None for value in values[:6].to_list())
|
||||
assert values[6] == pytest.approx(math.log(85.0))
|
||||
|
||||
|
||||
def test_ts_corr_two_series() -> None:
|
||||
values = _eval("ts_corr(close, volume, 3)")
|
||||
# 常数序列或完全单调: 只验证产出为有限值或 null, 无串组异常即可
|
||||
assert len(values) == 12
|
||||
|
||||
|
||||
# --------------------------------------------------------- 运行期 fail-closed
|
||||
|
||||
def test_runtime_missing_column_fails_closed() -> None:
|
||||
compiled = compile_formula("close * volume")
|
||||
assert compiled.ok
|
||||
frame_without_volume = _panel().drop("volume")
|
||||
assert compiled.frame_transform is not None
|
||||
assert compiled.frame_transform(frame_without_volume) is None # E013 语义
|
||||
assert compiled.frame_transform(_panel()) is not None
|
||||
@@ -0,0 +1,174 @@
|
||||
"""扩充批次 (2026-09-05) 16 个新虚拟因子的数值正确性测试。
|
||||
|
||||
合成单标的日频面板, 黄金值由 numpy 独立重算 (不复制实现),
|
||||
覆盖: 公式口径 / 无前视 (min_samples) / 除零 fail-closed / 列代数型因子。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date, timedelta
|
||||
|
||||
import numpy as np
|
||||
import polars as pl
|
||||
import pytest
|
||||
|
||||
from app.strategy.scoring import materialize_scoring_columns
|
||||
|
||||
N_DAYS = 250
|
||||
DATES = [date(2025, 1, 1) + timedelta(days=i) for i in range(N_DAYS)]
|
||||
T = np.arange(N_DAYS, dtype=float)
|
||||
|
||||
# 自然波动收益序列 (趋势 + 正弦): 避免等比价格的常数收益让波动率退化为浮点噪声
|
||||
DAILY_RET = 0.002 + 0.01 * np.sin(T / 9.0)
|
||||
CLOSE = 100.0 * np.cumprod(1.0 + DAILY_RET)
|
||||
OPEN = np.roll(CLOSE, 1) * 1.005
|
||||
OPEN[0] = 99.0 * 1.005 # 隔夜跳空 +0.5%
|
||||
PREV_CLOSE = np.roll(CLOSE, 1)
|
||||
PREV_CLOSE[0] = 99.0
|
||||
VOLUME = 1_000_000 + 500.0 * T # 量能缓增
|
||||
TURNOVER = VOLUME / 200_000_000.0 # 流通股本 2 亿股
|
||||
RET = np.concatenate([[np.nan], CLOSE[1:] / CLOSE[:-1] - 1.0])
|
||||
|
||||
# 列代数型因子的依赖列直接给黄金友好值
|
||||
RSI = 50.0 + 10.0 * np.sin(T / 7.0)
|
||||
MOM20 = np.concatenate([np.full(20, np.nan), CLOSE[20:] / CLOSE[:-20] - 1.0])
|
||||
MOM60 = np.concatenate([np.full(60, np.nan), CLOSE[60:] / CLOSE[:-60] - 1.0])
|
||||
KDJ_K = 50.0 + 15.0 * np.cos(T / 11.0)
|
||||
KDJ_D = 50.0 + 5.0 * np.sin(T / 13.0)
|
||||
AMPLITUDE = 0.02 + 0.0001 * T
|
||||
|
||||
|
||||
def _panel() -> pl.DataFrame:
|
||||
return pl.DataFrame({
|
||||
"symbol": ["TEST"] * N_DAYS,
|
||||
"date": DATES,
|
||||
"open": OPEN,
|
||||
"high": np.maximum(OPEN, CLOSE) * 1.01,
|
||||
"low": np.minimum(OPEN, CLOSE) * 0.99,
|
||||
"close": CLOSE,
|
||||
"prev_close": PREV_CLOSE,
|
||||
"volume": VOLUME,
|
||||
"amount": VOLUME * CLOSE * 100.0,
|
||||
"turnover_rate": TURNOVER,
|
||||
"rsi_14": RSI,
|
||||
"momentum_20d": MOM20,
|
||||
"momentum_60d": MOM60,
|
||||
"kdj_k": KDJ_K,
|
||||
"kdj_d": KDJ_D,
|
||||
"amplitude": AMPLITUDE,
|
||||
})
|
||||
|
||||
|
||||
def _col(frame: pl.DataFrame, name: str) -> np.ndarray:
|
||||
return frame[name].to_numpy()
|
||||
|
||||
|
||||
def _materialize(names: list[str]) -> pl.DataFrame:
|
||||
return materialize_scoring_columns(_panel(), names)
|
||||
|
||||
|
||||
def test_momentum_120d_golden() -> None:
|
||||
frame = _materialize(["momentum_120d"])
|
||||
got = _col(frame, "momentum_120d")
|
||||
golden = np.full(N_DAYS, np.nan)
|
||||
golden[120:] = CLOSE[120:] / CLOSE[:-120] - 1.0
|
||||
assert np.allclose(got[121:], golden[121:], atol=1e-12)
|
||||
assert np.isnan(got[:120]).all() # 无前视: 前 120 根必为空 (min_samples)
|
||||
|
||||
|
||||
def test_mom_accel_and_kdj_diff_column_algebra() -> None:
|
||||
frame = _materialize(["mom_accel_20_60", "kdj_kd_diff", "rsi_14_delta_5d"])
|
||||
assert np.allclose(_col(frame, "mom_accel_20_60"), MOM20 - MOM60, equal_nan=True)
|
||||
assert np.allclose(_col(frame, "kdj_kd_diff"), KDJ_K - KDJ_D, atol=1e-12)
|
||||
delta = np.full(N_DAYS, np.nan)
|
||||
delta[5:] = RSI[5:] - RSI[:-5]
|
||||
assert np.allclose(_col(frame, "rsi_14_delta_5d")[6:], delta[6:], atol=1e-12)
|
||||
|
||||
|
||||
def test_overnight_and_intraday_decomposition() -> None:
|
||||
frame = _materialize(["overnight_ret_20d", "intraday_ret_20d"])
|
||||
overnight_daily = OPEN / PREV_CLOSE - 1.0
|
||||
intraday_daily = CLOSE / OPEN - 1.0
|
||||
# 后向滚动窗: got[i] = sum(daily[i-19 .. i]); convolve('valid')[k] = sum(daily[k..k+19])
|
||||
# → got[i] 对应 valid[i-19], 从 i=21 起比对 (跳过合成首日 prev_close 特例)
|
||||
golden_on = np.convolve(overnight_daily, np.ones(20), "valid")[2:]
|
||||
golden_in = np.convolve(intraday_daily, np.ones(20), "valid")[2:]
|
||||
got_on = _col(frame, "overnight_ret_20d")
|
||||
got_in = _col(frame, "intraday_ret_20d")
|
||||
assert np.allclose(got_on[21:], golden_on, atol=1e-10)
|
||||
assert np.allclose(got_in[21:], golden_in, atol=1e-10)
|
||||
# 恒等式: sum(隔夜) + sum(日内) ≈ sum(全天收益); 精确差为每日交叉项 on*in
|
||||
# (跳空0.5% x 日内~1%, 20日累计 ~1e-3), 故用 5e-3 容差
|
||||
total = got_on[21:] + got_in[21:]
|
||||
golden_total = np.convolve(CLOSE / PREV_CLOSE - 1.0, np.ones(20), "valid")[2:]
|
||||
assert np.allclose(total, golden_total, atol=5e-3)
|
||||
|
||||
|
||||
def test_downside_vol_only_counts_negative_side() -> None:
|
||||
frame = _materialize(["downside_vol_20d"])
|
||||
got = _col(frame, "downside_vol_20d")[21:]
|
||||
for i, day in enumerate(range(21, N_DAYS)):
|
||||
window = np.minimum(RET[day - 19: day + 1], 0.0)
|
||||
golden = np.sqrt(np.mean(window ** 2))
|
||||
assert got[i] == pytest.approx(golden, abs=1e-12), f"day index {day}"
|
||||
|
||||
|
||||
def test_obv_trend_bounded_and_golden() -> None:
|
||||
frame = _materialize(["obv_trend_20d"])
|
||||
got = _col(frame, "obv_trend_20d")
|
||||
for day in range(21, N_DAYS, 25):
|
||||
window_ret = RET[day - 19: day + 1]
|
||||
window_vol = VOLUME[day - 19: day + 1]
|
||||
signed = np.sign(window_ret) * window_vol
|
||||
golden = signed.sum() / (window_vol.mean() * 20.0)
|
||||
assert got[day] == pytest.approx(golden, abs=1e-9), f"day index {day}"
|
||||
valid = got[~np.isnan(got)]
|
||||
assert (np.abs(valid) <= 1.0 + 1e-12).all() # 有界 [-1, 1]
|
||||
|
||||
|
||||
def test_log_float_mv_golden_and_fail_closed() -> None:
|
||||
frame = _materialize(["log_float_mv"])
|
||||
got = _col(frame, "log_float_mv")
|
||||
golden = np.log(CLOSE * VOLUME / TURNOVER)
|
||||
assert np.allclose(got, golden, atol=1e-10) # = ln(流通市值), 股本=2亿
|
||||
# 换手率为 0 → None (fail-closed, 不产生 inf)
|
||||
broken = _panel().with_columns(pl.lit(0.0).alias("turnover_rate"))
|
||||
out = materialize_scoring_columns(broken, ["log_float_mv"])
|
||||
assert out["log_float_mv"].is_null().all()
|
||||
|
||||
|
||||
def test_position_240d_and_distance_to_high() -> None:
|
||||
frame = _materialize(["position_240d", "distance_to_high_240d"])
|
||||
pos = _col(frame, "position_240d")
|
||||
dist = _col(frame, "distance_to_high_240d")
|
||||
for day in (241, 245, N_DAYS - 1):
|
||||
window = CLOSE[day - 239: day + 1]
|
||||
golden_pos = (CLOSE[day] - window.min()) / (window.max() - window.min())
|
||||
assert pos[day] == pytest.approx(golden_pos, abs=1e-12), f"pos day {day}"
|
||||
assert dist[day] == pytest.approx(CLOSE[day] / window.max() - 1.0, abs=1e-12)
|
||||
assert np.isnan(pos[:239]).all() # 无前视: 240 日窗在索引 239 才首次有效
|
||||
|
||||
|
||||
def test_vol_regime_amplitude_trend_turnover_stats() -> None:
|
||||
frame = _materialize(["vol_regime_5_60", "amplitude_trend_20_60", "turnover_mean_20d", "turnover_std_20d"])
|
||||
vr = _col(frame, "vol_regime_5_60")
|
||||
at = _col(frame, "amplitude_trend_20_60")
|
||||
tm = _col(frame, "turnover_mean_20d")
|
||||
ts = _col(frame, "turnover_std_20d")
|
||||
for day in (61, 120, N_DAYS - 1):
|
||||
fast = np.std(RET[day - 4: day + 1], ddof=1)
|
||||
slow = np.std(RET[day - 59: day + 1], ddof=1)
|
||||
assert vr[day] == pytest.approx(fast / slow, rel=1e-9, abs=1e-12), f"vr day {day}"
|
||||
a_fast = AMPLITUDE[day - 19: day + 1].mean()
|
||||
a_slow = AMPLITUDE[day - 59: day + 1].mean()
|
||||
assert at[day] == pytest.approx(a_fast / a_slow - 1.0, rel=1e-9, abs=1e-12)
|
||||
t_window = TURNOVER[day - 19: day + 1]
|
||||
assert tm[day] == pytest.approx(t_window.mean(), rel=1e-12)
|
||||
assert ts[day] == pytest.approx(t_window.std(ddof=1) / t_window.mean(), rel=1e-9)
|
||||
|
||||
|
||||
def test_amount_mean_20d_unit_is_yi() -> None:
|
||||
frame = _materialize(["amount_mean_20d"])
|
||||
got = _col(frame, "amount_mean_20d")
|
||||
day = N_DAYS - 1
|
||||
golden = (VOLUME[day - 19: day + 1] * CLOSE[day - 19: day + 1] * 100.0).mean() / 1e8
|
||||
assert got[day] == pytest.approx(golden, rel=1e-12)
|
||||
@@ -0,0 +1,315 @@
|
||||
"""因子注册表 (L-REG) P1 收口快照测试。
|
||||
|
||||
黄金数据为收口前 factor.py / scoring.py 的字面量副本。
|
||||
任何目录漂移 (id/label/group/desc/顺序/依赖/预热) 都必须在改动前更新这里的黄金数据,
|
||||
保证历史候选方案引用的因子 id 与挖掘调度顺序 (FACTOR_COLUMNS[:48]) 不受影响。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import pytest
|
||||
|
||||
from app.factors.registry import (
|
||||
FactorSpec,
|
||||
all_factors,
|
||||
factor_columns_view,
|
||||
factor_dependencies,
|
||||
get_factor,
|
||||
register_factor,
|
||||
scoring_warmups,
|
||||
virtual_dependencies,
|
||||
)
|
||||
|
||||
# --- 黄金数据: 收口前 factor.py FACTOR_COLUMNS 原文 ---
|
||||
GOLDEN_COLUMNS: list[dict] = [
|
||||
{"id": "momentum_5d", "label": "5日动量", "group": "动量", "desc": "5个交易日累计收益率"},
|
||||
{"id": "momentum_10d", "label": "10日动量", "group": "动量", "desc": "10个交易日累计收益率"},
|
||||
{"id": "momentum_20d", "label": "20日动量", "group": "动量", "desc": "20个交易日累计收益率"},
|
||||
{"id": "momentum_30d", "label": "30日动量", "group": "动量", "desc": "30个交易日累计收益率"},
|
||||
{"id": "momentum_60d", "label": "60日动量", "group": "动量", "desc": "60个交易日累计收益率"},
|
||||
{"id": "change_pct", "label": "日涨跌幅", "group": "动量", "desc": "当日收盘相对前收盘的收益率"},
|
||||
{"id": "ma5_bias", "label": "MA5乖离", "group": "均线偏离", "desc": "收盘价 / MA5 - 1"},
|
||||
{"id": "ma10_bias", "label": "MA10乖离", "group": "均线偏离", "desc": "收盘价 / MA10 - 1"},
|
||||
{"id": "ma20_bias", "label": "MA20乖离", "group": "均线偏离", "desc": "收盘价 / MA20 - 1"},
|
||||
{"id": "ma30_bias", "label": "MA30乖离", "group": "均线偏离", "desc": "收盘价 / MA30 - 1"},
|
||||
{"id": "ma60_bias", "label": "MA60乖离", "group": "均线偏离", "desc": "收盘价 / MA60 - 1"},
|
||||
{"id": "ema5_bias", "label": "EMA5乖离", "group": "均线偏离", "desc": "收盘价 / EMA5 - 1"},
|
||||
{"id": "ema10_bias", "label": "EMA10乖离", "group": "均线偏离", "desc": "收盘价 / EMA10 - 1"},
|
||||
{"id": "ema20_bias", "label": "EMA20乖离", "group": "均线偏离", "desc": "收盘价 / EMA20 - 1"},
|
||||
{"id": "ema30_bias", "label": "EMA30乖离", "group": "均线偏离", "desc": "收盘价 / EMA30 - 1"},
|
||||
{"id": "ema60_bias", "label": "EMA60乖离", "group": "均线偏离", "desc": "收盘价 / EMA60 - 1"},
|
||||
{"id": "rsi_6", "label": "RSI(6)", "group": "超买超卖", "desc": "6日相对强弱指标"},
|
||||
{"id": "rsi_14", "label": "RSI(14)", "group": "超买超卖", "desc": "14日相对强弱指标"},
|
||||
{"id": "rsi_24", "label": "RSI(24)", "group": "超买超卖", "desc": "24日相对强弱指标"},
|
||||
{"id": "macd_hist", "label": "MACD柱(原值)", "group": "趋势", "desc": "兼容历史研究; 跨股票比较建议优先使用MACD柱强度"},
|
||||
{"id": "macd_dif_pct", "label": "MACD DIF强度", "group": "趋势", "desc": "MACD DIF / 收盘价"},
|
||||
{"id": "macd_dea_pct", "label": "MACD DEA强度", "group": "趋势", "desc": "MACD DEA / 收盘价"},
|
||||
{"id": "macd_hist_pct", "label": "MACD柱强度", "group": "趋势", "desc": "MACD柱 / 收盘价, 消除股价尺度影响"},
|
||||
{"id": "kdj_k", "label": "KDJ-K", "group": "趋势", "desc": "KDJ指标K值"},
|
||||
{"id": "kdj_d", "label": "KDJ-D", "group": "趋势", "desc": "KDJ指标D值"},
|
||||
{"id": "kdj_j", "label": "KDJ-J", "group": "趋势", "desc": "KDJ指标J值"},
|
||||
{"id": "boll_position", "label": "布林位置", "group": "趋势", "desc": "收盘价在布林带下轨到上轨之间的位置"},
|
||||
{"id": "annual_vol_20d", "label": "20日波动率", "group": "波动率", "desc": "20日收益率年化标准差"},
|
||||
{"id": "atr_14", "label": "ATR(14)原值", "group": "波动率", "desc": "兼容历史研究; 跨股票比较建议优先使用ATR相对波动"},
|
||||
{"id": "atr_pct", "label": "ATR相对波动", "group": "波动率", "desc": "ATR(14) / 收盘价"},
|
||||
{"id": "amplitude", "label": "日振幅", "group": "波动率", "desc": "当日高低价差 / 前收盘价"},
|
||||
{"id": "boll_width", "label": "布林带宽", "group": "波动率", "desc": "布林带上下轨宽度 / MA20"},
|
||||
{"id": "vol_ratio_5d", "label": "5日量比", "group": "量价", "desc": "当日成交量 / 前5日平均成交量"},
|
||||
{"id": "vol_ratio_10d", "label": "10日量比", "group": "量价", "desc": "当日成交量 / 前10日平均成交量"},
|
||||
{"id": "vol_trend_5_10", "label": "成交量趋势", "group": "量价", "desc": "5日平均成交量 / 10日平均成交量 - 1"},
|
||||
{"id": "turnover_rate", "label": "换手率", "group": "量价", "desc": "使用历史时点流通股本计算的当日换手率"},
|
||||
{"id": "turnover_ratio_5d", "label": "换手率放大", "group": "量价", "desc": "当日换手率 / 前5日平均换手率 - 1"},
|
||||
{"id": "log_amount", "label": "成交额对数", "group": "量价", "desc": "ln(成交额 + 1), 降低极端规模影响"},
|
||||
{"id": "amount_ratio_5d", "label": "成交额放大", "group": "量价", "desc": "当日成交额 / 前5日平均成交额 - 1"},
|
||||
{"id": "gap_return", "label": "开盘跳空", "group": "价格位置", "desc": "开盘价 / 前收盘价 - 1"},
|
||||
{"id": "intraday_return", "label": "日内收益", "group": "价格位置", "desc": "收盘价 / 开盘价 - 1"},
|
||||
{"id": "close_position", "label": "收盘位置", "group": "价格位置", "desc": "收盘价在当日最低价到最高价之间的位置"},
|
||||
{"id": "distance_to_high_60d", "label": "距60日高点", "group": "价格位置", "desc": "收盘价 / 60日最高收盘价 - 1"},
|
||||
{"id": "distance_from_low_60d", "label": "距60日低点", "group": "价格位置", "desc": "收盘价 / 60日最低收盘价 - 1"},
|
||||
{"id": "vwap_bias", "label": "VWAP乖离", "group": "价格位置", "desc": "收盘价 / 当日成交均价 - 1, 成交均价 = 成交额 / (成交量x100)"},
|
||||
{"id": "max_ret_20d", "label": "20日最大单日涨幅", "group": "收益形态", "desc": "近20个交易日单日涨幅最大值(彩票效应, 高值代表博彩型特征强)"},
|
||||
{"id": "ret_skew_20d", "label": "20日收益偏度", "group": "收益形态", "desc": "近20个交易日日收益偏度, 高值代表右偏(偶发大涨)"},
|
||||
{"id": "up_days_20d", "label": "20日上涨天数", "group": "收益形态", "desc": "近20个交易日中上涨天数(0~20)"},
|
||||
{"id": "amihud_20d", "label": "20日Amihud非流动性", "group": "流动性", "desc": "近20日平均 |日涨跌幅| / 成交额(亿元), 高值代表流动性差"},
|
||||
{"id": "turnover_z_60d", "label": "换手率60日z分", "group": "流动性", "desc": "(当日换手率 - 前60日均值) / 前60日标准差, 衡量换手异动"},
|
||||
{"id": "vol_price_corr_20d", "label": "20日量价相关", "group": "量价", "desc": "近20个交易日日涨跌幅与成交量的相关系数, 高值代表量价同向"},
|
||||
{"id": "vol_trend_5_60", "label": "量能趋势(5/60)", "group": "量价", "desc": "5日平均成交量 / 60日平均成交量 - 1"},
|
||||
{"id": "limit_up_count_20d", "label": "涨停基因(20日)", "group": "涨停基因", "desc": "近20个交易日涨停次数"},
|
||||
{"id": "limit_up_count_60d", "label": "涨停基因(60日)", "group": "涨停基因", "desc": "近60个交易日涨停次数"},
|
||||
{"id": "pb_latest", "label": "市净率(最新公告)", "group": "财务", "desc": "收盘价 / 最新已公告每股净资产; 无财务数据或公告前为空"},
|
||||
{"id": "roe_latest", "label": "ROE(最新公告)", "group": "财务", "desc": "最新已公告净资产收益率(%); 无财务数据或公告前为空"},
|
||||
{"id": "gross_margin_latest", "label": "毛利率(最新公告)", "group": "财务", "desc": "最新已公告销售毛利率(%)"},
|
||||
{"id": "net_margin_latest", "label": "净利率(最新公告)", "group": "财务", "desc": "最新已公告销售净利率(%)"},
|
||||
{"id": "revenue_yoy_latest", "label": "营收增速(最新公告)", "group": "财务", "desc": "最新已公告营业收入同比(%)"},
|
||||
{"id": "net_income_yoy_latest", "label": "净利增速(最新公告)", "group": "财务", "desc": "最新已公告归母净利润同比(%)"},
|
||||
{"id": "debt_ratio_latest", "label": "资产负债率(最新公告)", "group": "财务", "desc": "最新已公告资产负债率(%)"},
|
||||
# --- 扩充批次 (2026-09-05): 追加于目录尾部, 前 48 项挖掘调度顺序不变 ---
|
||||
{"id": "log_float_mv", "label": "流通市值对数", "group": "规模", "desc": "ln(收盘价 x 当日成交量 / 换手率), 由换手率反推流通股本, 高值代表大盘"},
|
||||
{"id": "momentum_120d", "label": "120日动量", "group": "动量", "desc": "120个交易日累计收益率 (中期动量, 与短窗口互补)"},
|
||||
{"id": "mom_accel_20_60", "label": "动量加速度", "group": "动量", "desc": "20日动量 - 60日动量, 衡量近期动量相对中期是否增强"},
|
||||
{"id": "rsi_14_delta_5d", "label": "RSI五日变化", "group": "超买超卖", "desc": "RSI(14) - 5日前的RSI(14), 衡量强弱指标的边际变化"},
|
||||
{"id": "overnight_ret_20d", "label": "20日隔夜收益", "group": "收益形态", "desc": "近20日累计隔夜收益(开盘价/前收盘-1求和), A股隔夜与日内收益的定价机制不同"},
|
||||
{"id": "intraday_ret_20d", "label": "20日日内收益", "group": "收益形态", "desc": "近20日累计日内收益(收盘价/开盘价-1求和), 与隔夜收益构成收益分解"},
|
||||
{"id": "downside_vol_20d", "label": "20日下行波动", "group": "波动率", "desc": "sqrt(近20日 min(日收益,0)^2 均值), 只度量下跌侧风险"},
|
||||
{"id": "vol_regime_5_60", "label": "波动率状态(5/60)", "group": "波动率", "desc": "5日收益标准差 / 60日收益标准差, 高值代表波动骤然放大"},
|
||||
{"id": "amplitude_trend_20_60", "label": "振幅趋势(20/60)", "group": "波动率", "desc": "20日平均振幅 / 60日平均振幅 - 1"},
|
||||
{"id": "obv_trend_20d", "label": "20日量能潮", "group": "量价", "desc": "近20日 sign(日收益)x成交量 之和 / (20日均量x20), 有界[-1,1], 净买入方向的一致性"},
|
||||
{"id": "amount_mean_20d", "label": "20日均成交额(亿)", "group": "量价", "desc": "近20日平均成交额(亿元), 规模/流动性水平量"},
|
||||
{"id": "turnover_mean_20d", "label": "20日均换手", "group": "流动性", "desc": "近20日平均换手率, A股经典低换手溢价因子"},
|
||||
{"id": "turnover_std_20d", "label": "20日换手波动", "group": "流动性", "desc": "近20日换手率标准差 / 均值 (变异系数), 衡量交易活跃的稳定性"},
|
||||
{"id": "position_240d", "label": "一年价格位置", "group": "价格位置", "desc": "收盘价在近240个交易日最低价到最高价之间的位置 (0~1)"},
|
||||
{"id": "distance_to_high_240d", "label": "距一年高点", "group": "价格位置", "desc": "收盘价 / 240日最高收盘价 - 1, 接近0代表贴近一年新高"},
|
||||
{"id": "kdj_kd_diff", "label": "KDJ K-D差", "group": "趋势", "desc": "KDJ K值 - D值, 正值代表快线在慢线上方"},
|
||||
]
|
||||
|
||||
GOLDEN_VIRTUAL_DEPS: dict[str, frozenset[str]] = {
|
||||
**{
|
||||
f"ma{period}_bias": frozenset({"close", f"ma{period}"})
|
||||
for period in (5, 10, 20, 30, 60)
|
||||
},
|
||||
**{
|
||||
f"ema{period}_bias": frozenset({"close", f"ema{period}"})
|
||||
for period in (5, 10, 20, 30, 60)
|
||||
},
|
||||
"macd_dif_pct": frozenset({"close", "macd_dif"}),
|
||||
"macd_dea_pct": frozenset({"close", "macd_dea"}),
|
||||
"macd_hist_pct": frozenset({"close", "macd_hist"}),
|
||||
"boll_position": frozenset({"close", "boll_upper", "boll_lower"}),
|
||||
"atr_pct": frozenset({"close", "atr_14"}),
|
||||
"boll_width": frozenset({"ma20", "boll_upper", "boll_lower"}),
|
||||
"vol_ratio_10d": frozenset({"volume"}),
|
||||
"vol_trend_5_10": frozenset({"vol_ma5", "vol_ma10"}),
|
||||
"turnover_ratio_5d": frozenset({"turnover_rate"}),
|
||||
"log_amount": frozenset({"amount"}),
|
||||
"amount_ratio_5d": frozenset({"amount"}),
|
||||
"gap_return": frozenset({"open", "prev_close"}),
|
||||
"intraday_return": frozenset({"open", "close"}),
|
||||
"close_position": frozenset({"high", "low", "close"}),
|
||||
"distance_to_high_60d": frozenset({"close", "high_60d"}),
|
||||
"distance_from_low_60d": frozenset({"close", "low_60d"}),
|
||||
"max_ret_20d": frozenset({"close"}),
|
||||
"ret_skew_20d": frozenset({"close"}),
|
||||
"up_days_20d": frozenset({"close"}),
|
||||
"amihud_20d": frozenset({"close", "amount"}),
|
||||
"turnover_z_60d": frozenset({"turnover_rate"}),
|
||||
"vol_price_corr_20d": frozenset({"close", "volume"}),
|
||||
"vwap_bias": frozenset({"close", "volume", "amount"}),
|
||||
"vol_trend_5_60": frozenset({"volume"}),
|
||||
"limit_up_count_20d": frozenset({"consecutive_limit_ups"}),
|
||||
"limit_up_count_60d": frozenset({"consecutive_limit_ups"}),
|
||||
# --- 扩充批次 (2026-09-05) ---
|
||||
"log_float_mv": frozenset({"close", "volume", "turnover_rate"}),
|
||||
"momentum_120d": frozenset({"close"}),
|
||||
"mom_accel_20_60": frozenset({"momentum_20d", "momentum_60d"}),
|
||||
"rsi_14_delta_5d": frozenset({"rsi_14"}),
|
||||
"overnight_ret_20d": frozenset({"open", "prev_close"}),
|
||||
"intraday_ret_20d": frozenset({"open", "close"}),
|
||||
"downside_vol_20d": frozenset({"close"}),
|
||||
"vol_regime_5_60": frozenset({"close"}),
|
||||
"amplitude_trend_20_60": frozenset({"amplitude"}),
|
||||
"obv_trend_20d": frozenset({"close", "volume"}),
|
||||
"amount_mean_20d": frozenset({"amount"}),
|
||||
"turnover_mean_20d": frozenset({"turnover_rate"}),
|
||||
"turnover_std_20d": frozenset({"turnover_rate"}),
|
||||
"position_240d": frozenset({"close"}),
|
||||
"distance_to_high_240d": frozenset({"close"}),
|
||||
"kdj_kd_diff": frozenset({"kdj_k", "kdj_d"}),
|
||||
}
|
||||
|
||||
GOLDEN_WARMUP: dict[str, int] = {
|
||||
"vol_ratio_10d": 11,
|
||||
"turnover_ratio_5d": 6,
|
||||
"amount_ratio_5d": 6,
|
||||
"max_ret_20d": 21,
|
||||
"ret_skew_20d": 21,
|
||||
"up_days_20d": 21,
|
||||
"amihud_20d": 21,
|
||||
"turnover_z_60d": 61,
|
||||
"vol_price_corr_20d": 21,
|
||||
"vol_trend_5_60": 60,
|
||||
"limit_up_count_20d": 21,
|
||||
"limit_up_count_60d": 61,
|
||||
# --- 扩充批次 (2026-09-05) ---
|
||||
"momentum_120d": 121,
|
||||
"rsi_14_delta_5d": 6,
|
||||
"overnight_ret_20d": 21,
|
||||
"intraday_ret_20d": 21,
|
||||
"downside_vol_20d": 21,
|
||||
"vol_regime_5_60": 61,
|
||||
"amplitude_trend_20_60": 61,
|
||||
"obv_trend_20d": 21,
|
||||
"amount_mean_20d": 21,
|
||||
"turnover_mean_20d": 21,
|
||||
"turnover_std_20d": 21,
|
||||
"position_240d": 241,
|
||||
"distance_to_high_240d": 241,
|
||||
}
|
||||
|
||||
|
||||
def test_factor_columns_snapshot() -> None:
|
||||
"""注册表生成的 FACTOR_COLUMNS 与收口前字面量逐项一致 (含顺序)。"""
|
||||
from app.backtest.factor import FACTOR_COLUMNS
|
||||
|
||||
assert FACTOR_COLUMNS == GOLDEN_COLUMNS
|
||||
assert factor_columns_view() == GOLDEN_COLUMNS
|
||||
|
||||
|
||||
def test_virtual_dependencies_snapshot() -> None:
|
||||
"""注册表生成的依赖声明与收口前字面量逐项一致。"""
|
||||
from app.strategy.scoring import VIRTUAL_SCORING_DEPENDENCIES
|
||||
|
||||
assert VIRTUAL_SCORING_DEPENDENCIES == GOLDEN_VIRTUAL_DEPS
|
||||
assert virtual_dependencies() == GOLDEN_VIRTUAL_DEPS
|
||||
|
||||
|
||||
def test_scoring_warmup_snapshot() -> None:
|
||||
from app.strategy.scoring import _ROLLING_SCORING_WARMUP
|
||||
|
||||
assert _ROLLING_SCORING_WARMUP == GOLDEN_WARMUP
|
||||
assert scoring_warmups() == GOLDEN_WARMUP
|
||||
|
||||
|
||||
def test_catalog_counts_and_kinds() -> None:
|
||||
specs = all_factors()
|
||||
assert len(specs) == 77
|
||||
assert len({spec.id for spec in specs}) == 77 # id 唯一
|
||||
virtual = [spec for spec in specs if spec.kind == "virtual"]
|
||||
assert len(virtual) == 52 # ma/ema 10 + 原有 26 + 扩充批次 16
|
||||
financial = [spec for spec in specs if spec.pit]
|
||||
assert len(financial) == 7
|
||||
assert all(spec.pit_source == "financial_announce" for spec in financial)
|
||||
assert all(spec.asset_types == frozenset({"stock"}) for spec in financial)
|
||||
|
||||
|
||||
def test_mining_schedule_order_prefix_unchanged() -> None:
|
||||
"""挖掘调度取 FACTOR_COLUMNS[:48], 首元素必须保持 momentum_5d。"""
|
||||
from app.backtest.factor import FACTOR_COLUMNS
|
||||
|
||||
assert FACTOR_COLUMNS[0]["id"] == "momentum_5d"
|
||||
assert len(FACTOR_COLUMNS) >= 48
|
||||
|
||||
|
||||
def test_get_factor_and_dependencies() -> None:
|
||||
spec = get_factor("ma20_bias")
|
||||
assert spec is not None
|
||||
assert spec.dependencies == frozenset({"close", "ma20"})
|
||||
assert spec.warmup_bars == 1 # 无滚动窗口, 与历史默认一致
|
||||
|
||||
resolved = factor_dependencies(["ma20_bias", "rsi_14", "unknown_col"])
|
||||
assert resolved == frozenset({"close", "ma20", "rsi_14", "unknown_col"})
|
||||
|
||||
|
||||
def test_asset_type_filter() -> None:
|
||||
stock = all_factors(asset_type="stock")
|
||||
etf = all_factors(asset_type="etf")
|
||||
assert len(stock) == 77
|
||||
assert len(etf) == 70 # 财务 7 项仅股票
|
||||
|
||||
|
||||
def test_register_factor_rejects_duplicate() -> None:
|
||||
spec = get_factor("rsi_14")
|
||||
assert spec is not None
|
||||
with pytest.raises(ValueError, match="已注册"):
|
||||
register_factor(spec)
|
||||
|
||||
|
||||
def test_register_factor_allows_version_bump() -> None:
|
||||
from app.factors import registry
|
||||
|
||||
fresh = FactorSpec(id="__test_custom_factor", label="测试因子", group="测试", formula_text="close", kind="custom")
|
||||
register_factor(fresh)
|
||||
bumped = FactorSpec(
|
||||
id="__test_custom_factor", label="测试因子", group="测试", formula_text="close + 1",
|
||||
kind="custom", version=2,
|
||||
)
|
||||
register_factor(bumped)
|
||||
try:
|
||||
current = get_factor("__test_custom_factor")
|
||||
assert current is not None
|
||||
assert current.version == 2
|
||||
assert current.formula_text == "close + 1"
|
||||
finally:
|
||||
# 清理测试注册项; 目录视图 (_CATALOG) 不受 _REGISTRY 动态注册影响
|
||||
registry._REGISTRY.pop("__test_custom_factor", None)
|
||||
|
||||
|
||||
def _client():
|
||||
from fastapi import FastAPI
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from app.api.factors import router
|
||||
|
||||
app = FastAPI()
|
||||
app.include_router(router)
|
||||
return TestClient(app)
|
||||
|
||||
|
||||
def test_factors_api_contract() -> None:
|
||||
client = _client()
|
||||
response = client.get("/api/factors")
|
||||
assert response.status_code == 200
|
||||
payload = response.json()
|
||||
factors = payload["factors"]
|
||||
assert len(factors) == 77
|
||||
first = factors[0]
|
||||
assert first["id"] == "momentum_5d"
|
||||
assert first["kind"] == "base"
|
||||
assert first["formula"] == "5个交易日累计收益率"
|
||||
assert first["asset_types"] == ["etf", "stock"]
|
||||
ma20 = next(item for item in factors if item["id"] == "ma20_bias")
|
||||
assert ma20["kind"] == "virtual"
|
||||
assert ma20["dependencies"] == ["close", "ma20"]
|
||||
pb = next(item for item in factors if item["id"] == "pb_latest")
|
||||
assert pb["pit"] is True
|
||||
assert pb["asset_types"] == ["stock"]
|
||||
mv = next(item for item in factors if item["id"] == "log_float_mv")
|
||||
assert mv["kind"] == "virtual"
|
||||
assert mv["scale_free"] is False
|
||||
|
||||
|
||||
def test_factors_api_asset_filter_and_validation() -> None:
|
||||
client = _client()
|
||||
etf = client.get("/api/factors", params={"asset_type": "etf"}).json()["factors"]
|
||||
assert len(etf) == 70
|
||||
assert all("stock" in item["asset_types"] for item in etf)
|
||||
# 非法资产类型 → 422 (fail-closed, 不静默回退全量)
|
||||
assert client.get("/api/factors", params={"asset_type": "index"}).status_code == 422
|
||||
@@ -0,0 +1,148 @@
|
||||
"""自定义/复合因子存储与 scoring 桥测试 (P3)。"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import date
|
||||
|
||||
import polars as pl
|
||||
import pytest
|
||||
|
||||
from app.factors import store
|
||||
from app.factors.registry import (
|
||||
FactorSpec,
|
||||
all_factors,
|
||||
factor_columns_view,
|
||||
get_factor,
|
||||
unregister_factor,
|
||||
)
|
||||
from app.strategy import scoring
|
||||
|
||||
|
||||
@pytest.fixture()
|
||||
def cleanup_registry():
|
||||
"""测试注册的自定义因子在用例后清理, 不污染全局注册表。"""
|
||||
before = set()
|
||||
yield before
|
||||
for fid in before:
|
||||
unregister_factor(fid)
|
||||
|
||||
|
||||
def _panel(n_days: int = 30) -> pl.DataFrame:
|
||||
rows = []
|
||||
volumes = {"A": 1000.0, "B": 3000.0, "C": 2000.0}
|
||||
for index in range(n_days):
|
||||
for symbol, close in (("A", 10.0 + index), ("B", 50.0 - index), ("C", 20.0 + index * 2)):
|
||||
rows.append({
|
||||
"symbol": symbol, "date": date(2026, 1, index + 1),
|
||||
"close": close, "volume": volumes[symbol] + index, "amount": (1000.0 + index) * close,
|
||||
})
|
||||
return pl.DataFrame(rows).sort(["symbol", "date"])
|
||||
|
||||
|
||||
def test_custom_factor_definition_roundtrip(tmp_path, cleanup_registry) -> None:
|
||||
definition = {
|
||||
"id": "uf_test_rev", "kind": "custom", "version": 1, "label": "测试反转",
|
||||
"group": "自定义", "formula": "rank(-ts_sum(close / ts_delay(close, 1) - 1, 5))",
|
||||
"description": "", "direction": "low", "status": "draft",
|
||||
}
|
||||
spec = store.register_definition(definition)
|
||||
cleanup_registry.add("uf_test_rev")
|
||||
assert spec.kind == "custom"
|
||||
assert "close" in spec.dependencies
|
||||
assert spec.warmup_bars >= 6
|
||||
|
||||
store.save_one(tmp_path, definition)
|
||||
loaded = store.load_all(tmp_path)
|
||||
assert len(loaded) == 1 and loaded[0]["id"] == "uf_test_rev"
|
||||
|
||||
# 目录视图与 all_factors 追加动态因子
|
||||
ids = [item["id"] for item in factor_columns_view()]
|
||||
assert ids[:77] == [item["id"] for item in factor_columns_view()[:77]]
|
||||
assert "uf_test_rev" in ids and ids.index("uf_test_rev") >= 77
|
||||
assert any(s.id == "uf_test_rev" for s in all_factors())
|
||||
|
||||
# 快照约束不受影响: 未注册动态因子时目录 = 77 内置
|
||||
unregister_factor("uf_test_rev")
|
||||
assert len(factor_columns_view()) == 77
|
||||
|
||||
|
||||
def test_custom_factor_invalid_rejected(cleanup_registry) -> None:
|
||||
with pytest.raises(ValueError, match="E005"):
|
||||
store.register_definition({
|
||||
"id": "uf_bad", "kind": "custom", "label": "坏因子",
|
||||
"formula": "ts_delay(close, -3)", "status": "draft",
|
||||
})
|
||||
with pytest.raises(ValueError, match="uf_"):
|
||||
store.register_definition({
|
||||
"id": "wrong_prefix", "kind": "custom", "label": "坏前缀",
|
||||
"formula": "close", "status": "draft",
|
||||
})
|
||||
|
||||
|
||||
def test_composite_definition_and_cycle_guard(cleanup_registry) -> None:
|
||||
definition = {
|
||||
"id": "cf_test_combo", "kind": "composite", "version": 1, "label": "测试组合",
|
||||
"members": {"momentum_20d": 0.6, "turnover_rate": 0.4}, "status": "draft",
|
||||
}
|
||||
spec = store.register_definition(definition)
|
||||
cleanup_registry.add("cf_test_combo")
|
||||
assert spec.kind == "composite"
|
||||
assert spec.components == (("momentum_20d", 0.6), ("turnover_rate", 0.4))
|
||||
assert spec.dependencies == frozenset({"momentum_20d", "turnover_rate"})
|
||||
|
||||
# 自引用拒绝
|
||||
with pytest.raises(ValueError, match="自身"):
|
||||
store.to_spec({**definition, "id": "cf_self", "members": {"cf_self": 1.0, "close": 1.0}})
|
||||
|
||||
|
||||
def test_scoring_bridge_composite(cleanup_registry) -> None:
|
||||
"""复合因子经 scoring 物化: 截面加权 z 分可计算且依赖展开正确。"""
|
||||
store.register_definition({
|
||||
"id": "cf_ztest", "kind": "composite", "version": 1, "label": "桥接测试",
|
||||
"members": {"close": 0.5, "volume": 0.5}, "status": "active",
|
||||
})
|
||||
cleanup_registry.add("cf_ztest")
|
||||
|
||||
deps = scoring.scoring_dependencies({"cf_ztest": 1.0})
|
||||
assert deps == {"close", "volume"}
|
||||
assert scoring.scoring_warmup_bars({"cf_ztest": 1.0}) >= 1
|
||||
|
||||
frame = scoring.materialize_scoring_columns(_panel(), {"cf_ztest"})
|
||||
assert "cf_ztest" in frame.columns
|
||||
values = frame.filter(pl.col("date") == date(2026, 1, 10))["cf_ztest"]
|
||||
assert values.is_not_null().all()
|
||||
# 截面 z 之和的均值近似为 0 (等权两成员)
|
||||
assert abs(values.mean()) < 1e-9
|
||||
|
||||
|
||||
def test_scoring_bridge_custom_materializes(cleanup_registry) -> None:
|
||||
"""自定义 DSL 因子经 materialize_scoring_columns 物化 (与检验共用路径)。"""
|
||||
store.register_definition({
|
||||
"id": "uf_rank_close", "kind": "custom", "version": 1, "label": "价格排名",
|
||||
"formula": "rank(close)", "status": "draft",
|
||||
})
|
||||
cleanup_registry.add("uf_rank_close")
|
||||
frame = scoring.materialize_scoring_columns(_panel(), {"uf_rank_close"})
|
||||
assert "uf_rank_close" in frame.columns
|
||||
day = frame.filter(pl.col("date") == date(2026, 1, 1))
|
||||
assert day["uf_rank_close"].is_not_null().all()
|
||||
|
||||
|
||||
def test_load_into_registry_isolated_failure(tmp_path, cleanup_registry) -> None:
|
||||
good = {
|
||||
"id": "uf_good", "kind": "custom", "version": 1, "label": "好因子",
|
||||
"formula": "close + 1", "status": "draft",
|
||||
}
|
||||
store.save_one(tmp_path, good)
|
||||
(tmp_path / "user_data" / "custom_factors" / "uf_broken.json").write_text(
|
||||
"{ not json", encoding="utf-8"
|
||||
)
|
||||
loaded = store.load_into_registry(tmp_path)
|
||||
assert loaded == ["uf_good"]
|
||||
cleanup_registry.add("uf_good")
|
||||
|
||||
|
||||
def test_unregister_builtin_rejected() -> None:
|
||||
with pytest.raises(ValueError, match="内置"):
|
||||
unregister_factor("rsi_14")
|
||||
spec = get_factor("rsi_14")
|
||||
assert isinstance(spec, FactorSpec)
|
||||
@@ -0,0 +1,149 @@
|
||||
"""final 定版确认回归: 午休/收盘定版必须校验快照时间戳, 未达边界不落盘。
|
||||
|
||||
实测 (2026-09-04): 收盘定版在 15:00:02 仅凭"拉取成功"即标记完成并落盘, 但
|
||||
实时源当时仍返回 14:59:5x 的竞价前快照 (海鸥住工 7.07 而非官方收盘 7.10),
|
||||
旧价被永久固化到当日分区; 且重启后的盘后手动刷新会再次写回旧价。修复后:
|
||||
|
||||
- _process_full_market_records 收到 final_boundary_ms 时, 快照最大时间戳
|
||||
达到边界 (含容差) 才落盘/评估监控, 否则只更新展示缓存;
|
||||
- _final_boundary_ms/_past_final_deadline 提供边界与重试窗口 (收盘 15:00/15:30)。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import datetime, time as dt_time
|
||||
|
||||
import pytest
|
||||
|
||||
import app.services.quote_service as qs_module
|
||||
from app.market_time import CN_TZ, cn_today
|
||||
from app.services.quote_service import QuoteService
|
||||
|
||||
|
||||
def _beijing_ms(h: int, m: int, s: int = 0) -> int:
|
||||
return int(datetime.combine(cn_today(), dt_time(h, m, s), tzinfo=CN_TZ).timestamp() * 1000)
|
||||
|
||||
|
||||
def _record(ts_ms: int) -> dict:
|
||||
return {
|
||||
"symbol": "002084.SZ",
|
||||
"last_price": 7.07,
|
||||
"open": 7.00, "high": 7.10, "low": 6.95,
|
||||
"volume": 100_000, "amount": 707_000.0,
|
||||
"timestamp": ts_ms,
|
||||
}
|
||||
|
||||
|
||||
class _StubRepo:
|
||||
"""记录写盘调用的最小仓库桩。"""
|
||||
|
||||
def __init__(self) -> None:
|
||||
self.calls: list[str] = []
|
||||
|
||||
def get_index_symbol_set(self) -> set:
|
||||
return set()
|
||||
|
||||
def get_etf_instruments(self):
|
||||
import polars as pl
|
||||
return pl.DataFrame()
|
||||
|
||||
def flush_live_daily(self, df) -> None:
|
||||
self.calls.append("daily")
|
||||
|
||||
def flush_live_daily_asset(self, asset_type: str, df) -> None:
|
||||
self.calls.append(f"daily:{asset_type}")
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
def service(monkeypatch) -> tuple[QuoteService, _StubRepo, dict]:
|
||||
qs = QuoteService()
|
||||
repo = _StubRepo()
|
||||
qs._repo = repo
|
||||
events: dict[str, int] = {"broadcast": 0, "enriched": 0}
|
||||
monkeypatch.setattr(qs_module, "_persist_last_fetch", lambda ms: None)
|
||||
monkeypatch.setattr(qs, "_update_volume_delta", lambda *a, **k: None)
|
||||
monkeypatch.setattr(qs, "_evaluate_monitors", lambda *a, **k: None)
|
||||
monkeypatch.setattr(qs, "_broadcast_quote_updated", lambda: events.__setitem__("broadcast", events["broadcast"] + 1))
|
||||
monkeypatch.setattr(qs, "_flush_live_enriched", lambda *a, **k: events.__setitem__("enriched", events["enriched"] + 1))
|
||||
return qs, repo, events
|
||||
|
||||
|
||||
def test_final_snapshot_before_boundary_skips_disk(service) -> None:
|
||||
"""竞价前快照 (时间戳 < 15:00): 只更新展示缓存, 不写 daily/enriched。"""
|
||||
qs, repo, events = service
|
||||
boundary = _beijing_ms(15, 0)
|
||||
|
||||
qs._process_full_market_records(
|
||||
[_record(boundary - 60_000)], t0=0.0, now_ts=0.0,
|
||||
final_boundary_ms=boundary,
|
||||
)
|
||||
|
||||
assert qs._last_final_confirmed is False
|
||||
assert repo.calls == [] # 未写 kline_daily
|
||||
assert events["enriched"] == 0 # 未写 enriched
|
||||
assert events["broadcast"] == 1 # 展示缓存路径仍走通
|
||||
|
||||
|
||||
def test_final_snapshot_after_boundary_writes(service) -> None:
|
||||
"""边界后快照 (时间戳 ≥ 15:00): 定版落盘。"""
|
||||
qs, repo, events = service
|
||||
boundary = _beijing_ms(15, 0)
|
||||
|
||||
qs._process_full_market_records(
|
||||
[_record(boundary + 30_000)], t0=0.0, now_ts=0.0,
|
||||
final_boundary_ms=boundary,
|
||||
)
|
||||
|
||||
assert qs._last_final_confirmed is True
|
||||
assert repo.calls == ["daily"]
|
||||
assert events["enriched"] == 1
|
||||
assert events["broadcast"] == 1
|
||||
|
||||
|
||||
def test_snapshot_without_timestamp_never_confirmed(service) -> None:
|
||||
"""无时间戳的快照无法确认定版 → 不落盘 (交由盘后管道兜底)。"""
|
||||
qs, repo, events = service
|
||||
rec = _record(0)
|
||||
rec.pop("timestamp")
|
||||
|
||||
qs._process_full_market_records(
|
||||
[rec], t0=0.0, now_ts=0.0, final_boundary_ms=_beijing_ms(15, 0),
|
||||
)
|
||||
|
||||
assert qs._last_final_confirmed is False
|
||||
assert repo.calls == []
|
||||
assert events["enriched"] == 0
|
||||
|
||||
|
||||
def test_normal_poll_ignores_boundary(service) -> None:
|
||||
"""普通轮询 (无 final_boundary_ms): 时间戳在边界前也照常落盘。"""
|
||||
qs, repo, events = service
|
||||
|
||||
qs._process_full_market_records(
|
||||
[_record(_beijing_ms(14, 59))], t0=0.0, now_ts=0.0,
|
||||
)
|
||||
|
||||
assert qs._last_final_confirmed is None
|
||||
assert repo.calls == ["daily"]
|
||||
assert events["enriched"] == 1
|
||||
|
||||
|
||||
def test_final_boundary_ms_matches_beijing_close(monkeypatch) -> None:
|
||||
fake_now = datetime.combine(cn_today(), dt_time(15, 10), tzinfo=CN_TZ)
|
||||
monkeypatch.setattr(qs_module, "cn_now", lambda: fake_now)
|
||||
assert QuoteService._final_boundary_ms("close_final") == _beijing_ms(15, 0)
|
||||
assert QuoteService._final_boundary_ms("morning_final") == _beijing_ms(11, 30)
|
||||
assert QuoteService._final_boundary_ms("afternoon") is None
|
||||
|
||||
|
||||
def test_past_final_deadline(monkeypatch) -> None:
|
||||
def _at(h: int, m: int):
|
||||
return datetime.combine(cn_today(), dt_time(h, m), tzinfo=CN_TZ)
|
||||
|
||||
monkeypatch.setattr(qs_module, "cn_now", lambda: _at(15, 29))
|
||||
assert QuoteService._past_final_deadline("close_final") is False
|
||||
monkeypatch.setattr(qs_module, "cn_now", lambda: _at(15, 30))
|
||||
assert QuoteService._past_final_deadline("close_final") is True
|
||||
monkeypatch.setattr(qs_module, "cn_now", lambda: _at(12, 9))
|
||||
assert QuoteService._past_final_deadline("morning_final") is False
|
||||
monkeypatch.setattr(qs_module, "cn_now", lambda: _at(12, 11))
|
||||
assert QuoteService._past_final_deadline("morning_final") is True
|
||||
@@ -0,0 +1,110 @@
|
||||
"""metrics_v2 统计函数测试 (P3) — 黄金参考向量 + 性质断言。
|
||||
|
||||
NW t 的黄金值由测试内的独立第二实现 (显式循环求 Bartlett 加权长方差) 推导,
|
||||
与 stats_v2 向量化实现互为对拍。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import math
|
||||
|
||||
import numpy as np
|
||||
import pytest
|
||||
|
||||
from app.backtest.stats_v2 import (
|
||||
_normal_ppf,
|
||||
bh_fdr_qvalues,
|
||||
deflated_sharpe_psr,
|
||||
expected_max_sharpe,
|
||||
naive_t,
|
||||
newey_west_t,
|
||||
normal_two_sided_p,
|
||||
)
|
||||
|
||||
|
||||
def _nw_t_reference(values: list[float], lag: int) -> float:
|
||||
"""独立第二实现: 显式循环按定义计算 Bartlett 核 HAC t 值。"""
|
||||
n = len(values)
|
||||
mean = sum(values) / n
|
||||
centered = [value - mean for value in values]
|
||||
gamma = [
|
||||
sum(centered[i] * centered[i + lag_i] for i in range(n - lag_i)) / n
|
||||
for lag_i in range(lag + 1)
|
||||
]
|
||||
long_var = gamma[0]
|
||||
for lag_i in range(1, lag + 1):
|
||||
long_var += 2.0 * (1.0 - lag_i / (lag + 1)) * gamma[lag_i]
|
||||
se = math.sqrt(long_var / n)
|
||||
return mean / se
|
||||
|
||||
|
||||
def test_newey_west_matches_reference() -> None:
|
||||
rng = np.random.default_rng(42)
|
||||
values = np.cumsum(rng.normal(0, 0.01, 60)).tolist() # 高自相关
|
||||
for lag in (1, 3, 5):
|
||||
result = newey_west_t(values, lag)
|
||||
assert result is not None
|
||||
t_stat, mean, se = result
|
||||
assert t_stat == pytest.approx(_nw_t_reference(values, lag), rel=1e-9)
|
||||
assert mean == pytest.approx(float(np.mean(values)))
|
||||
assert se > 0
|
||||
|
||||
|
||||
def test_newey_west_deflates_autocorrelated_t() -> None:
|
||||
# 强正自相关序列: NW t 的绝对值必须小于朴素 t (自相关被正确惩罚)
|
||||
rng = np.random.default_rng(7)
|
||||
phi = 0.9
|
||||
values, last = [], 0.0
|
||||
for shock in rng.normal(0, 0.01, 500):
|
||||
last = phi * last + shock
|
||||
values.append(last)
|
||||
t_naive = naive_t(values)
|
||||
result = newey_west_t(values, lag=5)
|
||||
assert t_naive is not None and result is not None
|
||||
assert abs(result[0]) < abs(t_naive)
|
||||
|
||||
|
||||
def test_newey_west_insufficient_samples() -> None:
|
||||
assert newey_west_t([0.1, 0.2], lag=1) is None
|
||||
assert newey_west_t([], lag=1) is None
|
||||
assert newey_west_t([1.0] * 20, lag=1) is None # 零方差
|
||||
|
||||
|
||||
def test_bh_fdr_golden() -> None:
|
||||
# 经典 BH 示例 (Wikipedia): q = [.005, .02, .042, .042, .042]
|
||||
pvalues = [0.001, 0.008, 0.039, 0.041, 0.042]
|
||||
assert bh_fdr_qvalues(pvalues) == pytest.approx([0.005, 0.02, 0.042, 0.042, 0.042])
|
||||
# 乱序输入: q 值跟随原位置 (m=3: .042→r3 raw .042; .001→.003; .039→min(.0585,.042)=.042)
|
||||
assert bh_fdr_qvalues([0.042, 0.001, 0.039]) == pytest.approx([0.042, 0.003, 0.042])
|
||||
# None 透传
|
||||
assert bh_fdr_qvalues([None, 0.05]) == [None, 0.05]
|
||||
|
||||
|
||||
def test_normal_p_and_ppf_inverse() -> None:
|
||||
assert normal_two_sided_p(1.959964) == pytest.approx(0.05, abs=1e-6)
|
||||
assert normal_two_sided_p(0.0) == pytest.approx(1.0)
|
||||
assert _normal_ppf(0.975) == pytest.approx(1.959964, abs=1e-6)
|
||||
assert _normal_ppf(0.5) == pytest.approx(0.0, abs=1e-9)
|
||||
with pytest.raises(ValueError):
|
||||
_normal_ppf(0.0)
|
||||
|
||||
|
||||
def test_expected_max_sharpe_monotone() -> None:
|
||||
assert expected_max_sharpe(1, 0.04) == 0.0 # 单试验不校正
|
||||
assert expected_max_sharpe(10, 0.0) == 0.0
|
||||
# 试验数越多期望最大夏普越高 (越难超越)
|
||||
em_10 = expected_max_sharpe(10, 0.04)
|
||||
em_100 = expected_max_sharpe(100, 0.04)
|
||||
assert 0 < em_10 < em_100
|
||||
|
||||
|
||||
def test_deflated_sharpe_psr() -> None:
|
||||
# 无偏斜无超额峰度时退化为 Φ(SR * sqrt(n-1))
|
||||
probability = deflated_sharpe_psr(sharpe=0.1, n_obs=2500, skewness=0.0, kurtosis=3.0, expected_max_sharpe=0.0)
|
||||
assert probability == pytest.approx(0.5 * (1 + math.erf(0.1 * math.sqrt(2499) / math.sqrt(2))))
|
||||
# 校正项抬高分母会降低 PSR
|
||||
penalized = deflated_sharpe_psr(0.1, 2500, skewness=0.0, kurtosis=10.0)
|
||||
assert penalized < probability
|
||||
# EM 校正降低显著性
|
||||
deflated = deflated_sharpe_psr(0.1, 2500, 0.0, 3.0, expected_max_sharpe=0.08)
|
||||
assert deflated < probability
|
||||
assert deflated_sharpe_psr(0.1, 3) is None # 样本不足
|
||||
@@ -0,0 +1,403 @@
|
||||
# 因子平台总体方案（研究线产品与架构设计 v1）
|
||||
|
||||
> 状态标记约定与 `factor-system-design.md` 一致：
|
||||
> 【现状】= 当前仓库已存在、已核对代码；【设计】= 本方案提出、尚未实现。
|
||||
> 所有阶段代号 P0–P4 见 §10 路线图。
|
||||
|
||||
## 执行状态（2026-09-04）
|
||||
|
||||
P0–P3 已全部实现并通过验证（后端 98 项因子相关测试 + 全量套件 + 前端 build + 真实数据 E2E）：
|
||||
|
||||
- **P0 已实现**：`/factors` 一级页（检验/因子库/编辑器/组合 四 tab）、检验页 UX 六项、回测页 tab 移除与跳转、候选方案注入策略页。
|
||||
- **跟进整合（2026-09-04）**：侧边栏「挖掘」菜单移除，AI 挖掘并入 `/factors?tab=mining`（第 5 个 tab）；`/mining` 与 `/backtest?tab=mining` 旧链接保留参数重定向。
|
||||
- **自动挖掘（2026-09-05，P3 后续）**：检验页新增「自动挖掘」入口（`POST /api/backtest/mining/auto`）——L1 全量统计筛选（`app/services/auto_mining.py`，门槛按探索/均衡/严格三档）自动生成达标因子池，随请求持久化筛选摘要（达标清单 + 未达标原因分布），L2 相关性剪枝 / L3 束搜索组合 / L4 嵌套样本外验证与达标门槛全部复用现有挖掘引擎；worker 子进程启动时加载因子注册表（自定义/复合因子可入池）。组合搜索本身不依赖 LLM；AI 的位置留给后续"从失败模式生成新 DSL 因子假设"。
|
||||
- **因子 ↔ 策略结合（2026-09-05，四条桥全部已实现）**：① 策略触发器编辑内 Zap 快捷建因子条件信号（`SignalTriggerActions` + `AddFactorSignalDialog`，阈值给建议值，`csg_f_*` 自动挂接当前策略入场/卖出触发器）；② 因子库「生成策略」一键产出单因子排名策略（`GenerateFactorStrategyDialog` 渲染 `FactorRankResearchMatrixStrategy` 模板，matrix_native，保存为 `custom_factor_*`）；③ 自定义信号 AI 提示词注入全部因子分组（`custom_signals_ai.py::_format_fields`，行前缀「因子·」，白名单 `allowed_fields()` 收口）；④ 策略回测因子归因（`strategy.py` 入场候选因子快照 → 胜/败单入场信号日因子均值，结果区第 4 个 tab，`meta.scoring` 非空即产出；矩阵单因子排名类策略因 META.scoring 设计为空，暂不产出归因，为已知边界）。
|
||||
- **编辑器交互重构（2026-09-05）**：算子区合并为单一可点面板（全部 25 个 DSL 算子按「时序/截面/工具」分组，点击插入光标处，悬停显示签名与参数约束，替代"16 高频 chips + 静态速查表"双列表）；可用字段 chip 双语（英文 id 等宽 + 中文标签浅色）；「从模板开始」下拉新增「我的因子」分组（导入已保存自定义因子公式），非初始公式替换前弹确认防误覆盖。
|
||||
- **因子库扩充（2026-09-05）**：目录 61 → 77（新增 16 个虚拟因子，追加于目录尾部，挖掘调度前 48 项顺序不变）：规模（流通市值对数，由换手率反推流通股本）、收益分解（隔夜/日内 20 日累计）、120 日动量与动量加速度、RSI 边际变化、下行波动、波动率状态(5/60)、振幅趋势、量能潮 OBV、20 日均成交额、均换手/换手波动（低换手溢价）、一年价格位置与距一年高点、KDJ K-D 差。双内核同步实现（`strategy/scoring.py` 评分路径 + `backtest/matrix.py` 矩阵内核，ddof 口径对齐存量约定）；批量检验上限 64 → 96。
|
||||
- **P1 已实现**：`backend/app/factors/registry.py` 注册表（61 内置因子单一事实源）、快照/契约测试、`GET /api/factors`。
|
||||
- **P2 已实现**：`backend/app/factors/dsl.py` 公式编译器（词法→EBNF 语法→语义校验 E001–E016→Polars 表达式，两阶段求值解决跨截面套时序窗口）、`POST /api/factors/validate|trial`、编辑器 tab。
|
||||
- **P3 已实现**：`stats_v2.py`（Newey-West t / BH-FDR q / DSR，纯 numpy）、批量检验接线与前端服务端判读；`store.py` 自定义/复合因子持久化 + 评分桥（与策略共用同一条物化路径）+ CRUD API + 启动多轮加载；组合 tab；生命周期状态（草稿/激活/观察/退役）与状态流转 API。
|
||||
- **P4 部分实现**：生命周期状态已落地；**巡检 tab 与后台衰减定时任务延后**（见 §10），复合因子权重 ICIR 自动配置模式延后。
|
||||
- 验证环境备注：因子批量检验对 custom/composite 的底层依赖先经内置补算路径物化（`factor.py::_compute_missing_factors`），已按 fail-closed 语义处理缺失列。
|
||||
|
||||
## 0. 文档定位与阅读顺序
|
||||
|
||||
| 文档 | 角色 | 生效裁定 |
|
||||
| --- | --- | --- |
|
||||
| 本文（factor-platform-plan.md） | **产品与架构总体方案**：两线架构、导航与页面、交互设计、功能全集、路线图 | 本文 §2–§8 生效后，`factor-system-design.md` §11 的"不新增顶层页面"结论**由本文取代** |
|
||||
| `factor-system-design.md` | **后端技术 spec**：FactorSpec schema、DSL 文法与算子表、统计口径（NW/BH-FDR/DSR）、API 契约、缓存键、测试矩阵 | 除 §11 外全部继续有效；本文引用不复制 |
|
||||
|
||||
两文冲突时：产品/前端/导航以本文为准；后端 schema、统计公式、API 字段以旧文为准。
|
||||
|
||||
## 1. 架构总览：两线模型
|
||||
|
||||
### 1.1 模型
|
||||
|
||||
```text
|
||||
研究线（发现问题） 交易线（执行战法）
|
||||
┌─────────────────────────┐ ┌─────────────────────────┐
|
||||
│ 因子库(注册表) │ │ 策略引擎(战法) │
|
||||
│ ↑ 检验(IC/IR/多空) │ │ 时机/买卖规则/仓位/风控 │
|
||||
│ ↑ 公式DSL(自定义因子) │ │ ↑ 评分选股 │
|
||||
│ ↑ 组合因子(加权合成) │ │ ↑ 验证(鲁棒性/步进) │
|
||||
│ 挖掘(自动搜索因子组合) │ │ 回测(策略绩效) │
|
||||
└───────────┬─────────────┘ └───────────↑─────────────┘
|
||||
│ 唯一桥梁：策略 scoring 配置 │
|
||||
└────────── 因子(组合)权重 ──────────────────┘
|
||||
```
|
||||
|
||||
- **因子层**回答"此刻全市场谁排前面"（截面排序）；**战法层**回答"何时进、买几只、多重仓、何时出、怎么止损"。
|
||||
- 两线只通过**评分配置**结合：`{"scoring": {"cf_hotmom": 0.6, ...}}`。因子不会"变成"策略，因子给策略供货。
|
||||
- 【现状】桥已存在：`backend/app/strategy/scoring.py:13` `VIRTUAL_SCORING_DEPENDENCIES`（35 个虚拟因子依赖声明）、`scoring_value_expr`（按名取值表达式）、`mining.py:342` 挖掘候选产物本身就是 `scoring = dict(zip(factor_names, weights))`。
|
||||
- 【现状】研究线已分裂在两处：因子检验是 `/backtest?tab=factor`，挖掘是独立页 `/mining`（`frontend/src/router.tsx:126-127`）。本方案将研究线收拢。
|
||||
|
||||
### 1.2 各能力现状对照
|
||||
|
||||
| 能力 | 现状 | 本方案落点 |
|
||||
| --- | --- | --- |
|
||||
| 因子检验（批量/单因子） | 【现状】`backtest/factor.py` 61 因子、Rank IC/分年/衰减/分环境；前端 FactorDiscovery | §3（搬迁 + UX 重设计） |
|
||||
| 因子注册 | 【现状】硬编码 `FACTOR_COLUMNS`（`factor.py:36`），加因子=改代码 | §4 + 旧文 §2 L-REG 注册表 |
|
||||
| 公式配置 | 【现状】无（公式即代码） | §5 + 旧文 §3 DSL |
|
||||
| 自动挖掘 | 【现状】`mining.py` beam search + 嵌套验证 + 晋升门槛，`/mining` 页 | §7（协同设计，不搬家） |
|
||||
| 因子→策略 | 【现状】scoring 机制可用；复合因子需改代码 | §6 + 旧文 §8 FactorCompositeSpec |
|
||||
| 生命周期管理 | 【现状】无 | §8（P4） |
|
||||
|
||||
## 2. 信息架构与导航【设计】
|
||||
|
||||
### 2.1 导航变化
|
||||
|
||||
```text
|
||||
现状: 回测(/backtest: 因子|策略|验证) 挖掘(/mining) …
|
||||
目标: 回测(/backtest: 策略|验证) 因子(/factors) 挖掘(/mining) …
|
||||
↑ 导航顺序：回测 → 因子 → 挖掘 相邻
|
||||
```
|
||||
|
||||
- 新增一级路由 `/factors`，导航标签「因子」，插入在回测与挖掘之间（`router.tsx` + `Layout.tsx`，两者为二开文档 §5.2 高冲突热点，diff 必须最小）。
|
||||
- `/backtest` 瘦身为「策略」「验证」两个 tab。
|
||||
- **旧链接兼容**：`/backtest?tab=factor` → `<Navigate to="/factors" replace>`，沿用现有先例——`Backtest.tsx:38-43` 已用同一模式把 `tab=mining` 重定向到 `/mining` 并保留 query 参数。`tab=factor` 的重定向须保留 `run/candidate` 类参数的传递习惯（对齐 mining 先例）。
|
||||
|
||||
### 2.2 候选方案抽屉的归属
|
||||
|
||||
【现状】候选方案抽屉内联在 `Backtest.tsx:114`，头部门按钮打开；因子候选（FactorDiscovery「保存候选」）与策略候选共用该存储，且已有"载入复测"流转（`Backtest.tsx:35`）。
|
||||
|
||||
【设计】P0 将抽屉提取为共享组件 `CandidateDrawer`，**回测页与因子页都放入口**。裁定理由：候选是研究线的产物、交易线的原料，两边都要能开；提取是搬移不是复制（CONTRIBUTING §7 共享组件原则）。
|
||||
|
||||
### 2.3 因子页骨架
|
||||
|
||||
```text
|
||||
/factors
|
||||
├─ tab=inspect（检验） P0：现有 FactorDiscovery 整体迁入 + UX 六项改造（§3）
|
||||
├─ tab=library（因子库） P1：注册表浏览器（§4）
|
||||
├─ tab=editor（编辑器） P2：自定义公式因子（§5）
|
||||
├─ tab=composite（组合） P3：组合因子构建器（§6）
|
||||
└─ tab=lifecycle（巡检） P4：衰减监控看板（§8）
|
||||
```
|
||||
|
||||
- tab 状态走 URL query（`?tab=`），与回测页习惯一致，保证可收藏、可跳转。
|
||||
- P0 只落地 `inspect`，其余 tab 以"规划中"占位不渲染入口（YAGNI：不提前放空壳 tab）。
|
||||
|
||||
## 3. 检验 tab 详细设计（P0，纯前端）
|
||||
|
||||
### 3.1 页面线框
|
||||
|
||||
```text
|
||||
┌──────────────────────────────────────────────────────────────────────┐
|
||||
│ 因子检验 [候选方案] [AI 挖掘→] │
|
||||
│ ┌ 三句话说清楚（可折叠，默认展开，关后记忆到 localStorage）─────────┐ │
|
||||
│ │ ① 因子 = 用来给股票打分排序的特征（如 20日涨幅、换手率变化）。 │ │
|
||||
│ │ ② 这里检验：过去一段日子里，哪些特征真的能预测次日涨跌。 │ │
|
||||
│ │ ③ IC = 预测准确度（绝对值越大越准），IR = 稳定度（越大越稳）。 │ │
|
||||
│ └──────────────────────────────────────────────────────────────────┘ │
|
||||
│ ┌ 筛选配置 ────────────────────────────────────────────────────────┐ │
|
||||
│ │ 想好要测什么了吗？没有的话从预设开始： │ │
|
||||
│ │ [全面体检 61] [快速体检 10] [趋势动量 14] [超跌反转 9] │ │
|
||||
│ │ [量价资金 11] [财务价值 7] 当前: 全面体检 ✓（点预设=替换选择） │ │
|
||||
│ │ 已选 61/61 全选 清空 [🔍 搜索因子 (名称/公式)] │ │
|
||||
│ │ ┌─ 限高 45vh 内部滚动列表（现状保留）────────────────────────┐ │ │
|
||||
│ │ │ ▾ 动量 (6) 本组测：过去 N 日涨了多少 … │ │ │
|
||||
│ │ │ ☑ 5日动量 5个交易日累计收益率 │ │ │
|
||||
│ │ └────────────────────────────────────────────────────────────┘ │ │
|
||||
│ │ 资产与范围 / 日期 / 调仓 / 分组 / 佣金（现状保留） │ │
|
||||
│ │ [ 筛选 61 个因子 ] │ │
|
||||
│ └──────────────────────────────────────────────────────────────────┘ │
|
||||
│ ┌ 筛选结果 ─────────────────────────────────────────────────────────┐ │
|
||||
│ │ 结论句：61 个因子中 7 个有效*。最强：5日反转（IC −0.041）。 │ │
|
||||
│ │ 表格: 排名 | 因子(含方向箭头) | 预测力IC | 稳定度IR | IC胜率 | │ │
|
||||
│ │ 多空收益 | 最大回撤 | [结论徽章] | 操作 │ │
|
||||
│ └──────────────────────────────────────────────────────────────────┘ │
|
||||
└──────────────────────────────────────────────────────────────────────┘
|
||||
```
|
||||
|
||||
### 3.2 预设场景入口
|
||||
|
||||
**设计动机**：现状默认全选 61 个，用户第一个动作是"做选择题"；预设把第一个动作变成"按一下"。
|
||||
|
||||
预设清单（数量已按 `factor.py` FACTOR_COLUMNS 实际分组核对：动量 6、趋势 8、均线偏离 10、量价 9、价格位置 6、超买超卖 3、收益形态 3、波动率 5、流动性 2、涨停基因 2、财务 7，共 61）：
|
||||
|
||||
| 预设 | 数量 | 构成规则 | 适合谁 |
|
||||
| --- | --- | --- | --- |
|
||||
| 全面体检（默认） | 61 | 全部因子 | 不知道从哪开始的人；标注"耗时最长" |
|
||||
| 快速体检 | 10 | 除财务组外每组取窗口中位代表因子（如动量组取 20日动量） | 想快速看一圈的人 |
|
||||
| 趋势动量 | 14 | 动量组 + 趋势组 | 关心趋势行情下什么有效 |
|
||||
| 超跌反转 | 9 | 超买超卖组 + 价格位置组 | 关心抄底反转逻辑 |
|
||||
| 量价资金 | 11 | 量价组 + 流动性组 | 关心资金与量能信号 |
|
||||
| 财务价值 | 7 | 财务组 | 关心基本面（需财务数据源能力） |
|
||||
|
||||
交互规则：
|
||||
|
||||
1. 点预设 = **替换**当前选择集（不是叠加），按钮立即显示"当前预设 ✓"。
|
||||
2. 用户手动改勾选后，状态变为「自定义」，所有预设按钮取消高亮。
|
||||
3. 预设是**纯前端选择集**，不新增后端概念；构成规则在代码里以"组名 + 组内规则/显式 id 列表"声明，实现时从 `columns.data` 按组过滤生成（不硬编码 61 个 id 的第二份清单）。
|
||||
4. 「快速体检」的代表因子实现时从 FACTOR_COLUMNS 按"窗口中位"解析；解析失败的组整组纳入（fail-open 但数量标注会变）。
|
||||
|
||||
### 3.3 三句话引导条
|
||||
|
||||
- 位置：页面标题下方，结果区上方；`<details>` 式可折叠，默认展开，关闭状态记 `localStorage`（key 纳入现有 preferences 机制则更好，实现时择一，不新增后端字段）。
|
||||
- 文案即 §3.1 线框中三句，不许加第四句——解释超过三句就没人读。
|
||||
- 每句中的术语（IC/IR）与表头 tooltip（§3.4）文案保持一致。
|
||||
|
||||
### 3.4 表头判读 tooltip（文案定稿）
|
||||
|
||||
悬停表头即出，内容一句话 + 判读标准。列名同步通俗化（括号内保留原术语供专业用户）：
|
||||
|
||||
| 列 | 通俗列名 | tooltip 文案 |
|
||||
| --- | --- | --- |
|
||||
| IC 均值 | 预测力 IC | 每天用因子给股票打分、与次日真实涨跌算相关性（Rank IC）的均值。\|IC\|≥0.02 且稳定，即有预测力；负值同样有效（反向使用）。 |
|
||||
| IR | 稳定度 IR | IC 均值 ÷ IC 波动。≥0.3 值得关注，≥0.5 相当稳定。 |
|
||||
| IC 胜率 | 预测日占比 | IC 与预测方向一致的天数占比。50% 是抛硬币，55%+ 不错。 |
|
||||
| 多空收益 | 多空收益 | 每期买因子最高组、（模拟）卖最低组的累计收益差。注意 A 股做空受限，此列为理论口径。 |
|
||||
| 最大回撤 | 最大回撤 | 上述多空组合最痛的一段亏损幅度，衡量拿得住不住。 |
|
||||
| 排名 | 排名 | 按当前排序键排序的名次，默认按 \|IC\|。 |
|
||||
|
||||
### 3.5 结果判读（徽章 + 结论句）
|
||||
|
||||
- 【现状】结果行字段：`ic_mean / ir / ic_win_rate / long_short_return / long_short_max_drawdown / error`（`FactorDiscovery.tsx:372-376`）；**当前 API 无 t 值**。
|
||||
- 【设计·P0】客户端经验阈值判读（须在 UI 上明示"经验规则"）：
|
||||
- `有效`：`|ic_mean| ≥ 0.02 且 |ir| ≥ 0.3`
|
||||
- `边缘`：`|ic_mean| ≥ 0.02 或 |ir| ≥ 0.3`（其一成立）
|
||||
- `无效`：其余
|
||||
- `计算失败`：`error` 非空（现状已有 error 显示）
|
||||
- 【设计·P3 起】后端 metrics_v2（旧文 §6，`t_newey_west`）落地后，判读切换为**服务端显著性与阈值双条件**，前端仅渲染服务端 verdict；P0 的客户端阈值作为降级路径保留。
|
||||
- 结论句模板：`{N} 个因子中 {M} 个有效。最强：{label}（IC {值}）。`N=参与计算的因子数（排除 error），M=有效徽章数。结论句挂在表头右侧，字号小、颜色 muted。
|
||||
|
||||
### 3.6 因子方向标识
|
||||
|
||||
- 位置：**结果表**因子名旁，非复选框列表（检验前方向未知，不预测）。
|
||||
- 推导：`ic_mean > 0` → `↑ 值大看多`；`ic_mean < 0` → `↓ 值小看多`；null/error → 不显示。
|
||||
- 这是从数据推导的事实（样本内 IC 符号），非先验声明，因此**不进因子库元数据**（P1 因子库的方向列同样取最近一次检验的 IC 符号，并标注样本区间）。
|
||||
|
||||
### 3.7 路径闭环
|
||||
|
||||
1. **保存候选**后 toast：`已保存到候选方案（右上角查看）`；因子页头部有「候选方案」按钮（§2.2 共享抽屉）。
|
||||
2. **单因子检验**页（FactorDiscovery 内部 mode=single）顶部加返回链：`← 返回批量结果`，保留现有列表参数不丢。
|
||||
3. **因子库 → 检验**联动（P1）：`/factors?tab=inspect&focus={factor_id}` 打开检验页并只勾选该因子。
|
||||
4. **AI 挖掘 →**：因子页头部的 `/mining` 链接（研究线互链，§7）。
|
||||
|
||||
### 3.8 状态设计（CONTRIBUTING §7 强制项）
|
||||
|
||||
| 状态 | 触发 | 展示 |
|
||||
| --- | --- | --- |
|
||||
| 加载（因子清单） | columns.isLoading | 列表区骨架/「因子加载中…」（现状已有） |
|
||||
| 加载（运行中） | run.isPending | 按钮禁用 + 进度文案（现状已有），期间**保留上次结果**不闪空（现状行为核对） |
|
||||
| 空结果 | results=[] | EmptyState + 引导语「试试更长时间范围或减少因子」 |
|
||||
| 错误 | run.isError | 错误条 + 重试按钮，保留左侧配置 |
|
||||
| 无权限/无数据 | 财务预设但无财务数据源能力 | 预设按钮禁用 + title 说明「需要财务数据能力」（对齐能力矩阵门控习惯） |
|
||||
|
||||
### 3.9 迁移清单（P0 范围内的代码动作，供实施 PR 参考）
|
||||
|
||||
| # | 动作 | 文件 | 性质 |
|
||||
| --- | --- | --- | --- |
|
||||
| 1 | 新建 `/factors` 路由 + `Factors.tsx` 页壳（tab=inspect） | `router.tsx`、`pages/Factors.tsx` | 新增 |
|
||||
| 2 | FactorDiscovery 连同其 queryKeys 原样迁入 | `pages/backtest/FactorDiscovery.tsx` → 组件复用 | 搬移 |
|
||||
| 3 | `tab=factor` 重定向 `/factors`（保参，对齐 mining 先例） | `pages/Backtest.tsx` | 修改 |
|
||||
| 4 | Backtest 移除因子 tab，MODES 剩 策略/验证 | `pages/Backtest.tsx` | 修改 |
|
||||
| 5 | 候选抽屉提取共享组件，两页挂入口 | `Backtest.tsx` → `components/CandidateDrawer.tsx` | 提取 |
|
||||
| 6 | UX 六项（§3.2–§3.7） | `FactorDiscovery.tsx`（或其子组件拆分） | 新增 |
|
||||
| 7 | 导航项「因子」 | `Layout.tsx` | 修改 |
|
||||
|
||||
不做的：不改 queryKeys.ts 键结构、不改任何 API、不动 factor.py。
|
||||
|
||||
## 4. 因子库 tab（P1，依赖 L-REG 注册表）
|
||||
|
||||
### 4.1 列表
|
||||
|
||||
- 数据源：`GET /api/factors`（旧文 §10 已定义），P1 前无此接口时**本 tab 不上线**（不放假数据）。
|
||||
- 列：`id / 中文名 / 分组 / 类型（base｜virtual｜composite｜custom）/ 公式（desc 或 DSL）/ 最近 IC（含样本区间）/ 状态徽章 / 资产类型 / warmup 天数`。
|
||||
- 筛选：搜索框（id/名称/公式，复用检验 tab 同款逻辑）+ 类型、分组、状态下拉。
|
||||
- 行点击 → 详情抽屉。
|
||||
|
||||
### 4.2 详情抽屉
|
||||
|
||||
- FactorSpec 全字段（旧文 §2.2 的 17 字段）按人话展示：公式、方向（最近检验 IC 符号 + 样本区间）、warmup（"需要 X 日历史预热"）、依赖列、适用资产。
|
||||
- 操作：「去检验」（§3.7 联动）、「克隆到编辑器」（P2 后可用）、custom/composite 显示「编辑」「停用」。
|
||||
- 底部：最近一次检验报告摘要（IC/IR/多空/回撤）+「查看完整报告」跳检验 tab。
|
||||
|
||||
### 4.3 状态徽章
|
||||
|
||||
与 §8 状态机一致：`active / watch / retired / draft`。P1 只有 `active`（全部内置因子）与 `draft`（编辑器产物），watch/retired 在 P4 激活。
|
||||
|
||||
## 5. 因子编辑器 tab（P2，依赖 DSL 编译器）
|
||||
|
||||
### 5.1 布局
|
||||
|
||||
```text
|
||||
┌────────────────────────┬──────────────────────────┐
|
||||
│ 公式编辑区 │ 校验与预览区 │
|
||||
│ textarea (等宽字体) │ ✓/✗ 语法校验（错误码+定位） │
|
||||
│ 算子面板 (25 个, 三组) │ 依赖列集合 │
|
||||
│ 时序/截面/工具, 点击插入│ warmup = max(窗口) 提示 │
|
||||
│ 可用字段 ▾ (双语 chip) │ [试算最近 40 日 IC] │
|
||||
│ 从模板开始/我的因子 ▾ │ 试算结果: IC 走势 sparkline │
|
||||
│ [校验] [试算 40 日] │ 注册/更新表单 │
|
||||
└────────────────────────┴──────────────────────────┘
|
||||
```
|
||||
|
||||
### 5.2 交互流
|
||||
|
||||
1. 输入公式 → 调 `POST /api/factors/validate`（旧文 §10；错误码 E001–E016 见旧文 §3.5，含负 shift 即未来函数的 E005）。
|
||||
2. 校验通过 → 展示编译产物：依赖列集合、warmup_bars（旧文 §3 规则 4）。
|
||||
3. 试算：选近 40 交易日 + 当前资产，跑一次 Rank IC 快照（复用检验引擎计算路径，**不另写计算逻辑**）；展示 IC 序列 sparkline 与均值。
|
||||
4. 注册：状态 `draft`，写入注册表；后续在检验 tab 跑完整检验后转 `active`（§8 状态机）。编辑已有因子时公式变化保存为新版本并回草稿态。
|
||||
5. 模板起点：「从模板开始 / 我的因子」下拉——经典模板（5 个范式）+ 我的因子（已保存自定义因子公式导入）；替换非初始公式前弹确认。
|
||||
6. 点选插入：全部 25 个 DSL 算子按「时序(13)/截面(3)/工具(9)」分组，点击插入光标处（与后端 `dsl.OPERATORS` 同步维护），悬停显示签名与参数约束；79 个可用字段 chip 双语显示（英文 id 等宽 + 中文标签），点击插入。
|
||||
|
||||
### 5.3 约束（红线）
|
||||
|
||||
- 编辑器产物只经 DSL 编译到 Polars 表达式这一条计算路径（旧文 §3），**禁止**出现第二套求值器（CONTRIBUTING 禁止平行实现）。
|
||||
- 注册因子 id 冲突拒绝，不自增后缀静默改名。
|
||||
- 每一步的服务端校验 fail-closed：校验不过，注册按钮禁用且展示原因。
|
||||
|
||||
## 6. 组合 tab（P3，依赖 FactorCompositeSpec）
|
||||
|
||||
### 6.1 组合构建器
|
||||
|
||||
- 成分：从因子库多选（上限 8 个，对齐挖掘 max_factors 习惯），每行显示该因子最近 IC/IR。
|
||||
- 权重三模式：
|
||||
1. **手动**：逐项权重滑杆，实时归一显示；
|
||||
2. **ICIR 自动**：`w_i ∝ max(ICIR_i, 0)`，窗口 504 交易日，**只用 (t−window, t−1] 数据**（旧文 §8 点时规则，严禁 t 日数据参与 t 日权重）；
|
||||
3. **均等**：1/N。
|
||||
- 预览：选定回看区间内，组合 IC/IR 曲线 vs 各成分——证明"组合优于单因子"或及时劝退。
|
||||
|
||||
### 6.2 注册与桥接
|
||||
|
||||
- 「注册为复合因子」→ FactorCompositeSpec 入注册表，类型 `composite`。
|
||||
- 桥接机制**零引擎改动**（旧文 §8 已论证）：注册时自动写入 `VIRTUAL_SCORING_DEPENDENCIES` 对应条目 → 立即出现在策略评分可选列表。
|
||||
- 「一键创建策略」：打开策略新建表单并预填 `scoring = {组合id: 1.0}`，用户补战法参数——完成研究线→交易线的最后一步。
|
||||
|
||||
## 7. 与挖掘页协同【设计】
|
||||
|
||||
裁定：**挖掘保持独立页**（已有用户习惯 + 任务型界面本就该独立），做研究线互链而非合并。
|
||||
|
||||
1. 因子页头部「AI 挖掘 →」链到 `/mining`。
|
||||
2. 挖掘候选详情加「查看因子检验」：跳 `/factors?tab=inspect&focus=...` 预选该候选的成分因子。
|
||||
3. P4 可选：挖掘晋升的候选自动以 `watch` 状态写入因子库（人工确认后转 active）——默认不自动转正，晋升门槛（`mining.py` GATE_*）只证明样本外可行，不过不代表可持续。
|
||||
|
||||
## 8. 因子生命周期（P4）
|
||||
|
||||
### 8.1 状态机
|
||||
|
||||
```text
|
||||
draft ──(完整检验通过)──▶ active ──(衰减告警/人工)──▶ watch ──(持续衰减/人工)──▶ retired
|
||||
▲ │
|
||||
└──────────────(重新编辑公式, 重新走检验)──────────────────┘
|
||||
```
|
||||
|
||||
| 迁移 | 触发条件 | 动作 |
|
||||
| --- | --- | --- |
|
||||
| draft→active | 检验报告满足：\|IC\|≥0.02 且 \|IR\|≥0.3（或 metrics_v2 后 t 显著） | 进策略可选列表 |
|
||||
| active→watch | 滚动 63 日 IR < 0，或滚动 63 日 \|IC\| 较全样本减半 | 因子库标黄，不影响策略运行 |
|
||||
| watch→retired | 连续两个观察窗（126 日）仍衰减，或人工 | 从策略可选列表隐藏（已引用它的策略配置**不破坏**，仅新建不可选） |
|
||||
| retired→draft | 人工重新编辑 | 重走检验 |
|
||||
|
||||
### 8.2 衰减监控
|
||||
|
||||
- 计算：后台低频任务（日线收盘后），对 active/watch 因子重算滚动 63 日 Rank IC/IR——复用检验引擎同一条 IC 计算路径。
|
||||
- 展示：因子库 tab 状态列 + P4 的 `lifecycle` 看板（衰减榜：谁最近不行了）。
|
||||
- 通知：首版只做页面内徽章；接通知服务（`NotificationFormatter` 继承点）为可选后续，防打扰。
|
||||
|
||||
## 9. 后端支撑映射
|
||||
|
||||
| 本方案功能 | 后端依赖 | 旧文章节 | 代码落点 |
|
||||
| --- | --- | --- | --- |
|
||||
| 检验 tab（§3） | 无新依赖（P0 纯前端） | — | — |
|
||||
| 因子库（§4） | L-REG 注册表 + `GET /api/factors` | §2、§10 | 【设计】`app/factors/registry.py`（三处合一：factor.py:36 + scoring.py 虚拟因子 + pipeline ENRICHED_COLUMNS） |
|
||||
| 编辑器（§5） | DSL 编译器 + `POST /api/factors/validate` | §3、§10 | 【设计】`app/factors/dsl.py` |
|
||||
| 判读升级（§3.5） | metrics_v2（t_newey_west 等） | §6 | factor.py `_calc_ic` 扩展（现状 `factor.py:738-750`） |
|
||||
| 组合（§6） | FactorCompositeSpec + scoring 桥 | §8 | scoring.py `VIRTUAL_SCORING_DEPENDENCIES`（现状 `scoring.py:13`，写路径自动化） |
|
||||
| 生命周期（§8） | 衰减重算任务 + 状态字段 | §6（IC 口径复用） | 【设计】注册表状态字段 + 低频任务 |
|
||||
|
||||
## 10. 路线图（PR 切分）
|
||||
|
||||
> 与旧文 §14 的十个 PR 合并视图：数据线四个 PR（st_history / delisted_kline / industry_pit / index_ext）独立并行，不受本路线图阻塞。
|
||||
|
||||
### P0 — 因子页拆分 + 检验 UX（1 个 PR，纯前端）
|
||||
|
||||
- 范围：§2 导航/路由/重定向 + §3.9 迁移清单 7 项 + §3.2–§3.7 UX 六项。
|
||||
- 依赖：无。
|
||||
- 验收：`/factors` 可用；旧链接 301 语义跳转；候选抽屉两页可用；六项 UX 逐项截图；窄屏检查；`pnpm build` 过；queryKeys 无变更。
|
||||
- 回滚：revert 单 PR 即回到现状，无数据迁移。
|
||||
|
||||
### P1 — 注册表 + 因子库（2 个 PR：后端 / 前端）
|
||||
|
||||
- 后端：L-REG 注册表（旧文 §2，FACTOR_COLUMNS 由注册表生成、对外常量保留兼容别名；scoring.py 改读注册表、函数签名不变）+ `GET /api/factors`。
|
||||
- 前端：因子库 tab 上线（§4）。
|
||||
- 验收：注册表生成结果与现 61 因子逐项一致（快照测试）；旧候选方案引用的历史 id 全部可读。
|
||||
|
||||
### P2 — DSL + 编辑器（2 个 PR）
|
||||
|
||||
- 后端：DSL 编译器 + validate/试算端点（旧文 §3，含 E001–E016）。
|
||||
- 前端：编辑器 tab（§5）。
|
||||
- 验收：错误因子公式全路径测试；试算与检验引擎同源计算对拍一致。
|
||||
|
||||
### P3 — metrics_v2 + 组合 + 判读升级（2 个 PR）
|
||||
|
||||
- 后端 a：metrics_v2（NW t 值等，旧文 §6；测试=黄金参考向量 + 可选 dev 组 statsmodels 对拍）。
|
||||
- 后端 b+前端：FactorCompositeSpec + scoring 桥自动化 + 组合 tab（§6）+ 结果判读切服务端（§3.5）。
|
||||
- 验收：组合因子在策略评分中可选且数值路径唯一；DSR/BH-FDR 数值测试过。
|
||||
|
||||
### P4 — 生命周期（1 个 PR，可延后）
|
||||
|
||||
- §8 状态机 + 衰减任务 + 巡检看板。
|
||||
- 验收：衰减告警可复现（构造样本）；retired 因子不破坏存量策略配置（兼容测试）。
|
||||
|
||||
## 11. 验证矩阵（对齐 CONTRIBUTING §9）
|
||||
|
||||
| 改动 | 最低验证 |
|
||||
| --- | --- |
|
||||
| P0 前端 | `pnpm build`；手工检查：加载/空/错/禁用/无权限五态、预设切换与自定义态、搜索不丢选择（回归本轮已验证的行为）、旧链接重定向、候选抽屉双页、窄屏 |
|
||||
| P1 注册表 | 后端 pytest：注册表快照=现 61 因子、历史 id 兼容、scoring 行为不变（对拍测试）；API 契约测试 |
|
||||
| P2 DSL | 编译器单测（合法/非法公式全错误码）、防未来函数（E005 用例）、试算与检验同源对拍 |
|
||||
| P3 统计 | NW/BH-FDR/DSR 数值断言（黄金向量）；组合因子端到端：注册→策略评分可选→回测出数 |
|
||||
| P4 生命周期 | 衰减触发用例、retired 兼容用例、低频任务不阻塞启动 |
|
||||
| 全阶段 | `git diff --check`；每 PR 附界面证据（截图覆盖改动前后） |
|
||||
|
||||
## 12. 兼容性与风险
|
||||
|
||||
| 项 | 影响 | 对策 |
|
||||
| --- | --- | --- |
|
||||
| 旧链接 `/backtest?tab=factor` | 收藏/外链失效 | Navigate 重定向保参（mining 先例） |
|
||||
| 候选方案存储 | 无 schema 变化 | 仅抽屉组件提取，存储与"载入复测"链路不动 |
|
||||
| 因子 id | 不变 | P0 不碰 factor.py；P1 注册表生成需逐项快照对齐 |
|
||||
| `router.tsx`/`Layout.tsx` 高冲突热点 | 上游合并冲突面 | diff 最小化：只加一条路由/一个导航项，不动既有结构 |
|
||||
| P0 判读阈值是经验值 | 可能误判"有效" | UI 明示"经验规则"；P3 换服务端显著性后降级为兜底 |
|
||||
| 方向箭头取自样本内 IC 符号 | 过拟合误读 | tooltip 注明"样本内方向，历史不代表未来" |
|
||||
| 编辑器产生大量低质因子 | 注册表膨胀 | draft 不进策略可选列表；必须过完整检验才 active |
|
||||
|
||||
## 13. 不做什么（YAGNI，续旧文 §16）
|
||||
|
||||
1. 因子市场/分享/导入第三方因子包——无真实需求前不做。
|
||||
2. 因子自动再训练/自动调参流水线——自动权重窗口已是点时滚动，更细即过拟合。
|
||||
3. 实时因子流（盘中逐 tick 因子值）——研究线是日频截面，实时因子属于监控线的事。
|
||||
4. 跨市场（港美股）因子——数据源与交易规则均不支持。
|
||||
5. 因子血缘图谱可视化——注册表依赖列字段已可追溯，图形化为炫技。
|
||||
6. 把挖掘页合并进因子页——任务型界面独立存在（§7 裁定）。
|
||||
|
||||
## 14. 完成标准
|
||||
|
||||
本方案"完成"= 路线图 P0–P3 全部合入且：
|
||||
|
||||
1. `/factors` 四 tab（检验/库/编辑器/组合）可用，旧链接可达；
|
||||
2. 一个不懂量化的用户按 §3.1 引导能独立跑出一次检验并读懂结论句（可用性自测标准）；
|
||||
3. 一个自定义公式因子走完 编辑→校验→试算→注册→检验→进策略评分 全链路无断点；
|
||||
4. 全程未产生第二套计算逻辑（注册/编辑器/组合均编译到同一条 Polars/检验路径）；
|
||||
5. P0–P3 各 PR 验收标准逐项通过，界面证据与测试结果真实可查。
|
||||
@@ -0,0 +1,842 @@
|
||||
# 因子体系专业化设计（提案）
|
||||
|
||||
> **状态声明**：本文是设计提案，**尚未实现**。凡标注【现状】的条目引用当前仓库真实代码（基于 main@2ce8b4b1），可直接核对；凡标注【设计】的条目是目标契约，**不得当作已存在的 API 导入或调用**（遵循 `docs/secondary-development.md` 第 1 节的状态区分要求）。
|
||||
>
|
||||
> 全部【现状】引用已于 2026-09-04 逐条核对，Polars API 与依赖可行性已实测（polars 1.40.1），验证记录见 §17；一处初稿引用错误（pipeline.py:1738→1795）已修正。
|
||||
>
|
||||
> 涉及改动均按二次开发分级标注（L1 配置 / L2 扩展点 / L3 核心源码修改，见 `docs/secondary-development.md` 第 2 节）。
|
||||
|
||||
---
|
||||
|
||||
## 0. 设计目标
|
||||
|
||||
1. **因子定义单一事实源**：公式、元数据、计算、测试同处一地，可审计、可版本化。
|
||||
2. **研究结论可辩护**：宇宙可解释、风险调整显式、统计检验完备、指标口径唯一。
|
||||
3. **策略接入零摩擦**【用户核心诉求】：因子研究成果（含用户自定义因子、复合因子、挖掘产物）以统一形态被策略评分、选股、回测、监控四端消费，一处定义、处处生效。
|
||||
4. **不推倒重来**：挖掘框架（purge/embargo/嵌套样本外）、enriched 列体系、虚拟评分机制全部保留，只做补层和收口。
|
||||
|
||||
---
|
||||
|
||||
## 1. 分层总览与现状映射
|
||||
|
||||
| 层 | 目标模块 | 现状代码 | 动作 | 分级 |
|
||||
| --- | --- | --- | --- | --- |
|
||||
| L-REG 因子注册表 | `app/factors/registry.py`【设计】 | `backtest/factor.py:36` FACTOR_COLUMNS + `strategy/scoring.py` 虚拟因子 + `indicators/pipeline.py` ENRICHED_COLUMNS | 三处合一收口 | L3（重构） |
|
||||
| L-DSL 表达式层 | `app/factors/dsl/`【设计】 | 无（`strategy/custom_signals.py` 白名单模式可借鉴） | 新增 | L2 |
|
||||
| L-UNI 宇宙构建 | `app/factors/universe.py`【设计】 | 无（tradable/limit_up_locked/listing_date 素材已存在） | 新增 | L2 |
|
||||
| L-NEU 风险调整 | `app/factors/neutralize.py`【设计】 | 无（`get_index_daily`、行业 preset、`share_capital.py` 素材已存在） | 新增 | L2 |
|
||||
| L-INF 统计检验 | `app/factors/stats.py`【设计】 | 无 | 新增 | L2 |
|
||||
| L-MET 指标统一 | `app/factors/metrics.py`【设计】 | `backtest/engine.py` 三种 Sharpe（:2899/:2993/:3116） | 收敛 + 版本化 | L3（热点） |
|
||||
| L-CMP 复合因子→策略 | `app/factors/composite.py`【设计】 | `strategy/scoring.py` 虚拟评分字段机制【现状·已可用】 | 扩展既有机制 | L2→L3 接线 |
|
||||
| 数据契约 | provider dataset 声明 | 无 ST 历史/退市股/点时行业 | 新增 dataset | L1（YAML）+provider 实现 |
|
||||
|
||||
模块落点说明：新建 `app/factors/` 包而不是塞进 `backtest/`，因为因子目录、宇宙、中性化被选股（`strategy/`）、回测（`backtest/`）、挖掘（`backtest/mining.py`）三方消费，放任一方都会造成反向依赖（违反 CONTRIBUTING 2.3 模块边界）。
|
||||
|
||||
---
|
||||
|
||||
## 2. 因子注册表（L-REG）
|
||||
|
||||
### 2.1 现状问题
|
||||
|
||||
因子元数据目前分散在四处，互相漂移无感知：
|
||||
|
||||
| 位置 | 内容 | 缺陷 |
|
||||
| --- | --- | --- |
|
||||
| `backtest/factor.py:36-109` | 62+ 因子目录（id/label/group/desc） | desc 是自然语言，与计算无绑定 |
|
||||
| `strategy/scoring.py:13-51` | VIRTUAL_SCORING_DEPENDENCIES | 虚拟因子的依赖声明，但与因子目录是两套清单 |
|
||||
| `strategy/scoring.py:80+` | `scoring_value_expr` | 虚拟因子的 Polars 表达式，硬编码 if-else 分发 |
|
||||
| `strategy/scoring.py:53-66` | `_ROLLING_SCORING_WARMUP` | 预热窗口第三套清单 |
|
||||
|
||||
### 2.2 FactorSpec 完整 schema【设计】
|
||||
|
||||
```python
|
||||
@dataclass(frozen=True)
|
||||
class FactorSpec:
|
||||
id: str # 全局唯一，^f_[a-z0-9_]{1,40}$；内置因子保持现有列名不变（如 momentum_20d）
|
||||
version: int # 因子语义版本；公式变更必须 +1，进缓存键
|
||||
label: str # 中文显示名
|
||||
group: str # 展示分组（沿用现有：动量/均线偏离/超买超卖/趋势/波动率/量价/…）
|
||||
kind: Literal["base", "virtual", "composite", "custom"]
|
||||
# base: 已物化在 enriched parquet
|
||||
# virtual: 按需由 base 列编译计算（如 ma5_bias）
|
||||
# composite: 复合因子（见 §8）
|
||||
# custom: 用户 DSL 因子（见 §3）
|
||||
expr_factory: Callable[[frozenset[str]], pl.Expr | None] | None
|
||||
# virtual/custom 的计算：输入可用列集合，依赖不完整返回 None（fail-closed）
|
||||
formula_text: str # 人类可读公式；virtual 由表达式自动生成，base 手写并配特征化测试锁定
|
||||
dependencies: frozenset[str] # 展开到 enriched base 列（自递归展开 composite/custom 依赖）
|
||||
direction: Literal["high", "low", "none"]
|
||||
# 预期信号方向；进复合因子默认权重与 UI 排序展示
|
||||
unit: Literal["ratio", "pct", "score", "count", "days", "currency", "none"]
|
||||
# 单位口径，UI 格式化与 sanity check 用（禁止"数值<1 乘 100"启发式）
|
||||
warmup_bars: int # 历史窗口需求（交易日数）；对齐 _ROLLING_SCORING_WARMUP 语义
|
||||
pit: bool # 是否点时数据依赖（财务因子 = True）
|
||||
pit_source: Literal["financial_announce", "share_capital_announce", "none"]
|
||||
asset_types: frozenset[Literal["stock", "etf"]]
|
||||
incremental_safe: bool # 盘中增量路径（pipeline.py:1795）能否复算；False 则盘中不含该列
|
||||
scale_free: bool # 跨标的可比（可直接截面排序）；如 atr_14 原值 = False，atr_pct = True
|
||||
null_policy: Literal["keep", "drop_row"]
|
||||
# 研究路径默认 keep（不填零，沿用 fundamentals.py 纪律）
|
||||
stability: Literal["stable", "experimental", "deprecated"]
|
||||
tags: tuple[str, ...] # 风格标签："momentum"/"value"/"size"/"lottery"/"liquidity"/…
|
||||
```
|
||||
|
||||
注册表 API（仅内部 Python 接口，不新增 HTTP）：
|
||||
|
||||
```python
|
||||
register_factor(spec) # 启动期注册；重复 id 且 version 未增 → 拒绝启动（fail-closed）
|
||||
get_factor(fid) -> FactorSpec
|
||||
all_factors(asset_type=None, stable_only=False) -> list[FactorSpec]
|
||||
factor_dependencies(fids) -> frozenset[str] # 递归展开
|
||||
factor_value_exprs(available_cols) -> dict[str, pl.Expr | None]
|
||||
```
|
||||
|
||||
### 2.3 迁移策略【设计】
|
||||
|
||||
1. **特征化测试先行**：固定样本（≥50 只股票 × 含除权日、停牌日、涨跌停日的窗口）快照当前全部 62+ 因子在 enriched 与 `scoring_value_expr` 两条路径的输出，重构后断言逐位一致。扩展 `backend/tests/backtest/test_factor_library_v2.py`。
|
||||
2. `VIRTUAL_SCORING_DEPENDENCIES`、`scoring_value_expr` 的 if-else 分发、`_ROLLING_SCORING_WARMUP` 逐一改读注册表，**函数签名不变**（`scoring.py` 对外契约保持）。
|
||||
3. `FACTOR_COLUMNS` 改由注册表生成，`factor.py` 对外常量保留为兼容别名。
|
||||
4. desc 公式与 `formula_text` 不一致处，以特征化测试输出的实际计算为准修正文档。
|
||||
|
||||
### 2.4 因子分类学与补全清单【设计】
|
||||
|
||||
现有 11 组保留；补全以下专业常用因子（标注数据依赖，缺数据不注册、不静默）：
|
||||
|
||||
| 族 | 建议新增 | 公式要点 | 依赖 |
|
||||
| --- | --- | --- | --- |
|
||||
| 动量 | 特质动量 `f_idio_mom_20d` | 个股日收益对基准收益回归残差的 20 日累计 | 指数日K（已有） |
|
||||
| 动量 | 52 周新高接近度 `f_near_high_52w` | close / 250 日最高 close − 1 | 已有 |
|
||||
| 反转 | 短期反转 `f_rev_5d` | −momentum_5d（direction=low 的语义化封装） | 已有 |
|
||||
| 波动 | 已实现波动偏度差、高低频波动分解 | 简化：`f_vol_ratio_short_long` = vol_5d/vol_60d | 已有 |
|
||||
| 波动(条件) | 条件波动率 `f_ewma_vol` | RiskMetrics EWMA(λ=0.94) 条性日波动年化；比等权 rolling_std 对近端冲击响应更快，低成本低争议 | 已有（`ewm_std` 向量化） |
|
||||
| 波动(条件) | 波动的波动 `f_vol_of_vol_60d` / 波动区制 `f_vol_regime` | 波动率的滚动 std / EWMA 波动 ÷ 长期波动；区分"高波市场"与"波动突变"，A 股风格切换敏感因子 | 已有 |
|
||||
| 波动(条件) | GARCH(1,1) 条件波动 `f_garch_vol` | **标记 experimental、按需实现**：逐 symbol 递归拟合与全向量化管线冲突，若引入必须走 `numba_runtime.py`【现状】路径或 numba/arch 依赖，先以 EWMA 交付（日频下 EWMA ≈ GARCH 的 90% 价值） | 已有 + numba |
|
||||
| 量价 | 量价背离 `f_pv_divergence_20d` | −vol_price_corr_20d 语义化 | 已有 |
|
||||
| 流动性 | 非流动性变化 `f_amihud_chg` | amihud_20d / amihud_60d − 1 | 已有 |
|
||||
| 规模 | 流通市值对数 `f_log_float_mv` | ln(历史流通股本 × raw_close)【点时股本，share_capital.py 已有】 | 已有 |
|
||||
| 价值 | `f_ep_latest`、`f_ep_ttm` | 1/PE 口径（E/P 比 PE 统计性质更好）；ttm 需财务四表滚动 | 财务（已有）；ttm 需扩展 |
|
||||
| 质量 | 应收/存货增速差、商誉/净资产 | 财务表字段 | 财务（需字段核对） |
|
||||
| 涨停 | 首板/连板区分、炸板后回封率 | 基于 consecutive_limit_ups、炸板列 | 已有 |
|
||||
| 财务 | SUE（盈余惊喜） | (E_t − E_{t-4}) / σ(ΔE, 4期)，公告日口径 | 财务历史（已有 `_merge_report_history`） |
|
||||
|
||||
---
|
||||
|
||||
### 2.5 全量因子目录映射表（代码推导，PR-6 注册蓝本）
|
||||
|
||||
下表覆盖现有全部 61 个因子【现状：`factor.py:36-109`】，kind 与依赖由 `scoring.py:13-51` VIRTUAL_SCORING_DEPENDENCIES 逐字推导（virtual = 表中出现的键，base = 未出现即已物化列），运行时预热取自 `scoring.py:53-66` `_ROLLING_SCORING_WARMUP` 代码值。direction 列为**建议初值**（high=因子值大预期收益高；"待标定"= 振荡类/方向依市场状态，PR-6 注册时依 IC 实证方向标定并允许研究层覆盖）。
|
||||
|
||||
| 因子 | 组 | kind | 依赖（virtual 展开） | 预热 | direction |
|
||||
| --- | --- | --- | --- | --- | --- |
|
||||
| momentum_5d/10d/20d/30d/60d | 动量 | base | — | 全局120日 | high |
|
||||
| change_pct | 动量 | base | — | 全局 | high |
|
||||
| ma5..60_bias（5个） | 均线偏离 | virtual | {close, maN} | 全局 | high |
|
||||
| ema5..60_bias（5个） | 均线偏离 | virtual | {close, emaN} | 全局 | high |
|
||||
| rsi_6 / rsi_14 / rsi_24 | 超买超卖 | base | — | 全局 | 待标定 |
|
||||
| macd_hist | 趋势 | base | — | 全局 | 待标定 |
|
||||
| macd_dif_pct / macd_dea_pct / macd_hist_pct | 趋势 | virtual | {close, macd_dif/dea/hist} | 全局 | high |
|
||||
| kdj_k / kdj_d / kdj_j | 趋势 | base | — | 全局 | 待标定 |
|
||||
| boll_position | 趋势 | virtual | {close, boll_upper, boll_lower} | 全局 | high |
|
||||
| annual_vol_20d | 波动率 | base | — | 全局 | low |
|
||||
| atr_14 | 波动率 | base | — | 全局 | 待标定 |
|
||||
| atr_pct | 波动率 | virtual | {close, atr_14} | 全局 | low |
|
||||
| amplitude | 波动率 | base | — | 全局 | low |
|
||||
| boll_width | 波动率 | virtual | {ma20, boll_upper, boll_lower} | 全局 | low |
|
||||
| vol_ratio_5d | 量价 | base | — | 全局 | 待标定 |
|
||||
| vol_ratio_10d | 量价 | virtual | {volume} | 11 | 待标定 |
|
||||
| vol_trend_5_10 | 量价 | virtual | {vol_ma5, vol_ma10} | 全局 | high |
|
||||
| turnover_rate | 量价 | base | — | 全局 | low |
|
||||
| turnover_ratio_5d | 量价 | virtual | {turnover_rate} | 6 | high |
|
||||
| log_amount | 量价 | virtual | {amount} | 全局 | 待标定 |
|
||||
| amount_ratio_5d | 量价 | virtual | {amount} | 6 | high |
|
||||
| gap_return | 价格位置 | virtual | {open, prev_close} | 全局 | 待标定 |
|
||||
| intraday_return | 价格位置 | virtual | {open, close} | 全局 | 待标定 |
|
||||
| close_position | 价格位置 | virtual | {high, low, close} | 全局 | 待标定 |
|
||||
| distance_to_high_60d | 价格位置 | virtual | {close, high_60d} | 全局 | high |
|
||||
| distance_from_low_60d | 价格位置 | virtual | {close, low_60d} | 全局 | high |
|
||||
| vwap_bias | 价格位置 | virtual | {close, volume, amount} | 全局 | 待标定 |
|
||||
| max_ret_20d | 收益形态 | virtual | {close} | 21 | low |
|
||||
| ret_skew_20d | 收益形态 | virtual | {close} | 21 | low |
|
||||
| up_days_20d | 收益形态 | virtual | {close} | 21 | 待标定 |
|
||||
| amihud_20d | 流动性 | virtual | {close, amount} | 21 | low |
|
||||
| turnover_z_60d | 流动性 | virtual | {turnover_rate} | 61 | 待标定 |
|
||||
| vol_price_corr_20d | 量价 | virtual | {close, volume} | 21 | 待标定 |
|
||||
| vol_trend_5_60 | 量价 | virtual | {volume} | 60 | high |
|
||||
| limit_up_count_20d | 涨停基因 | virtual | {consecutive_limit_ups} | 21 | high |
|
||||
| limit_up_count_60d | 涨停基因 | virtual | {consecutive_limit_ups} | 61 | high |
|
||||
| pb_latest | 财务 | base(点时联表) | — | 公告日机制 | low |
|
||||
| roe_latest | 财务 | base(点时联表) | — | 公告日机制 | high |
|
||||
| gross_margin_latest | 财务 | base(点时联表) | — | 公告日机制 | high |
|
||||
| net_margin_latest | 财务 | base(点时联表) | — | 公告日机制 | high |
|
||||
| revenue_yoy_latest | 财务 | base(点时联表) | — | 公告日机制 | high |
|
||||
| net_income_yoy_latest | 财务 | base(点时联表) | — | 公告日机制 | high |
|
||||
| debt_ratio_latest | 财务 | base(点时联表) | — | 公告日机制 | low |
|
||||
|
||||
计数核对：virtual 35 + base 非财务 19 + 财务 7 = 61，与 FACTOR_COLUMNS 一致。base 因子的研究预热由 `FACTOR_WARMUP_DAYS=120`（`factor.py:111`【现状】）统一承担；财务因子 pit=true、pit_source=financial_announce。
|
||||
|
||||
## 3. 因子表达式层（L-DSL）
|
||||
|
||||
### 3.1 语法与算子表【设计】
|
||||
|
||||
表达式 = `expr ::= operand | expr op expr | func(expr[, expr[, const]])`；中缀 + 函数调用，无变量赋值、无循环。
|
||||
|
||||
**操作数**：基准列（open/high/low/close/volume/amount/turnover_rate/prev_close/raw_close）、白名单指标列（注册表中 base 因子）、已注册因子 id（virtual/composite/custom，递归内联展开）、数值常量。
|
||||
|
||||
**时序算子**（`over("symbol")`，窗口 n ∈ [2, 512]，全部只向后看）：
|
||||
|
||||
| 算子 | 语义 | Polars 编译 |
|
||||
| --- | --- | --- |
|
||||
| `ts_mean(x,n)` / `ts_std(x,n)` / `ts_sum(x,n)` | 滚动均值/样本标准差/求和 | `rolling_mean/std/sum(n)` |
|
||||
| `ts_max(x,n)` / `ts_min(x,n)` | 滚动极值 | `rolling_max(n)` / `rolling_min(n)` |
|
||||
| `ts_delta(x,n)` | x − ts_delay(x,n) | `x - x.shift(n)` |
|
||||
| `ts_delay(x,n)` | n 期前的值（n ∈ [1, 512]，**禁止负数**——负数即未来函数，编译期报错） | `x.shift(n)` |
|
||||
| `ts_rank(x,n)` | 当期值在滚动窗口内的分位 | `rolling_rank(n)`【已验证：polars 1.40.1 存在且行为正确，§17】 |
|
||||
| `ts_zscore(x,n)` | (x − ts_mean)/ts_std | 组合表达式 |
|
||||
| `ts_corr(x,y,n)` / `ts_cov(x,y,n)` | 滚动相关/协方差 | 顶层函数 `pl.rolling_corr(x,y,window_size=n)` / `pl.rolling_cov`【已验证：Expr 上无此方法，必须走顶层函数，§17】 |
|
||||
| `ts_quantile(x,n,q)` | 滚动分位（q ∈ (0,1) 常量） | `rolling_quantile` |
|
||||
| `decay_linear(x,n)` | 线性衰减加权均值（近端权重大） | 手写权重组合表达式 |
|
||||
|
||||
**v1 不提供的时序算子及原因**：`ts_argmax/ts_argmin`——Polars 无向量化实现（`rolling_map` 为逐窗 Python 回调，违反向量化约束，已验证 Expr 无 `rolling_arg_max`）；"距极值天数"类需求以具体因子的组合表达式实现（如 `distance_to_high_60d` 模式），确有高频需求再经 numba 扩展。
|
||||
|
||||
**截面算子**（按日期分组，逐日横截面）：
|
||||
|
||||
| 算子 | 语义 | 说明 |
|
||||
| --- | --- | --- |
|
||||
| `rank(x)` | 横截面百分位排名 ∈ (0,1] | null 不参与排名 |
|
||||
| `zscore(x)` | 横截面 (x−μ)/σ | σ=0 → null |
|
||||
| `winsorize(x,k)` | 截尾至 μ±kσ（k ∈ [1,6] 常量，默认 3） | 截面口径 |
|
||||
|
||||
**算术/工具**：`+ − * /`（除零 → null）、`log abs sign sqrt min max power(x,c) clamp(x,lo,hi)`、三元 `if_else(cond, a, b)`、比较与逻辑 `> >= < <= == != and or not`（产出布尔，配合 if_else）。
|
||||
|
||||
### 3.2 校验规则（编译期全部强制）【设计】
|
||||
|
||||
1. 标识符必须在基准列/白名单/已注册因子内，否则报错（防注入，沿用 `custom_signals.py` 白名单哲学）。
|
||||
2. `ts_delay`/`ts_delta` 的 n ≥ 0；任何窗口 n ∈ [2, 512]；`power` 指数 |c| ≤ 4；AST 深度 ≤ 12；表达式 token 数 ≤ 200。
|
||||
3. 常量折叠后若产生 `x/0` 类静态除零 → 编译失败。
|
||||
4. 依赖列集合 = 递归展开；warmup_bars = max(各 ts 算子窗口)；超出即注册表标记，研究 UI 提示所需历史长度。
|
||||
5. 产出类型必须为数值或布尔（布尔经 `cast` 视为 0/1）。
|
||||
6. **禁止未来引用的总闸**：所有时序算子 shift 语义已内建，语法层不存在负 shift；code review checklist 补一条"新增算子必须只向后看"。
|
||||
|
||||
### 3.3 编译流水线【设计】
|
||||
|
||||
`text → tokenizer → Pratt 解析 → AST → 语义检查(§3.2) → 依赖/预热推导 → Polars Expr 工厂`。产出缓存（表达式文本 → 编译产物 LRU，键含依赖列版本）；编译失败返回结构化错误（位置 + 原因），不抛裸异常。
|
||||
|
||||
### 3.4 形式文法（EBNF）与错误码目录【设计】
|
||||
|
||||
```ebnf
|
||||
expr = or_expr ;
|
||||
or_expr = and_expr { "or" and_expr } ;
|
||||
and_expr = cmp_expr { "and" cmp_expr } ;
|
||||
cmp_expr = add_expr [ (">" | ">=" | "<" | "<=" | "==" | "!=") add_expr ] ;
|
||||
add_expr = mul_expr { ("+" | "-") mul_expr } ;
|
||||
mul_expr = unary { ("*" | "/") unary } ;
|
||||
unary = "-" unary | primary ;
|
||||
primary = NUMBER | IDENT | func_call | "(" expr ")" ;
|
||||
func_call = IDENT "(" [ arglist ] ")" ;
|
||||
arglist = expr { "," expr } ;
|
||||
(* IDENT：基准列/白名单指标列/已注册因子 id/算子名；NUMBER：十进制与负号经 unary 处理 *)
|
||||
```
|
||||
|
||||
运算符优先级由产生式层级固定（or < and < 比较 < 加减 < 乘除 < 一元负号 < 原子），与 Python/JS 语义一致，降低用户迁移成本。
|
||||
|
||||
**错误码目录**（编译与运行校验的唯一错误词汇表，API/编辑器/UI 共用）：
|
||||
|
||||
| 码 | 含义 | 触发 |
|
||||
| --- | --- | --- |
|
||||
| E001 | 未知标识符 | IDENT 不在白名单/注册表 |
|
||||
| E002 | 未知函数 | 函数名不在算子表 |
|
||||
| E003 | 参数数量/类型不符 | 算子签名不匹配（含常量参数位置） |
|
||||
| E004 | 窗口越界 | n∉[2,512] 或 q∉(0,1) |
|
||||
| E005 | 负 shift | ts_delay/ts_delta 的 n<0 |
|
||||
| E006 | 嵌套深度超限 | AST 深度>12 |
|
||||
| E007 | 规模超限 | token 数>200 |
|
||||
| E008 | 静态除零 | 常量折叠检出分母恒 0 |
|
||||
| E009 | 产出类型非法 | 非数值/布尔 |
|
||||
| E010 | power 指数越界 | \|c\|>4 |
|
||||
| E011 | winsorize k 越界 | k∉[1,6] |
|
||||
| E012 | 循环引用 | 因子依赖成环（含自定义因子链） |
|
||||
| E013 | 依赖列不可用 | 面板缺列（运行时） |
|
||||
| E014 | 语法错误 | 解析失败（附位置） |
|
||||
| E015 | 预热不足 | warmup > 研究窗口（运行时） |
|
||||
| E016 | 常量表达式 | 无任何标识符，拒绝保存 |
|
||||
|
||||
错误响应统一结构：`{"code": "E001", "message": "未知标识符: clos", "position": {"offset": 12, "line": 1}, "detail": {...}}`。
|
||||
|
||||
### 3.5 用户因子生命周期与存储【设计】
|
||||
|
||||
- 存储路径：`data/user_data/custom_factors/*.json`（对齐 custom_signals 目录约定），schema：
|
||||
|
||||
```json
|
||||
{
|
||||
"id": "uf_my_rev", // ^uf_[a-z0-9_]{1,40}$，前缀与 csg_ 同哲学
|
||||
"version": 1,
|
||||
"label": "我的反转因子",
|
||||
"formula": "rank(-ts_sum(change_pct, 5))",
|
||||
"direction": "low",
|
||||
"description": "5 日累计涨幅的截面倒数",
|
||||
"created_at": "2026-09-04T00:00:00",
|
||||
"updated_at": "2026-09-04T00:00:00"
|
||||
}
|
||||
```
|
||||
|
||||
- 生命周期：草稿（编辑器内试算，不落盘）→ 保存（编译通过 + 试算有非空输出才可保存，fail-closed）→ 引用（策略 scoring / 因子研究 / 复合因子）→ 版本化（公式变更 version+1，旧结果按 version 键隔离）→ 删除（有引用时列出引用方并二次确认，对齐策略删除的 fail-closed 要求）。
|
||||
- 加载失败的单个文件只禁用该因子并提示，不影响启动与其他因子（对齐 plugins 隔离要求，CONTRIBUTING 第 4 节）。
|
||||
|
||||
---
|
||||
|
||||
## 4. 宇宙构建（L-UNI）
|
||||
|
||||
### 4.1 UniverseSpec 完整 schema【设计】
|
||||
|
||||
```python
|
||||
@dataclass(frozen=True)
|
||||
class UniverseSpec:
|
||||
exclude_suspended: bool = True # 停牌（tradable 矩阵口径，matrix.py:1476+）
|
||||
exclude_limit_locked: bool = True # 调仓时点一字涨停不可买入者（buy_limit_up 口径）
|
||||
exclude_st: bool = False # 非点时（今日名称），开启时报告中必须出现降级注记
|
||||
min_listing_days: int = 0 # 次新剔除；listing_date 已入库未使用（api/data.py:765）
|
||||
min_amount_quantile: float | None = None # 流动性过滤：当日成交额截面分位下限 (0,1)
|
||||
cap_quantile_range: tuple[float, float] | None = None # 市值分位区间；依赖 f_log_float_mv
|
||||
max_names: int | None = None # 截面数量上限；排序键由研究上下文显式传入（因子值或复合分，不隐式默认），用于微型宇宙研究
|
||||
```
|
||||
|
||||
### 4.2 执行语义【设计】
|
||||
|
||||
- 按日生成 `universe[date] -> set[symbol]`，**每个过滤条件独立短路、独立计数**，产出 `filter_stats`（每日各过滤器剔除数），研究报告展示"宇宙从 5200 → 4980 → 4890"漏斗。
|
||||
- `universe_id = sha256(canonical_json(spec))[:12]`，进一切下游缓存键。
|
||||
- 过滤顺序固定（先便宜的列过滤，后需联表的），顺序本身进 canonical_json。
|
||||
- **as_of 语义**：宇宙内一切判定只用当日及以前数据。ST 例外必须显式标注 `degraded: ["st_not_point_in_time"]` 并在报告 UI 渲染黄条。
|
||||
|
||||
### 4.3 数据缺口降级矩阵【设计】
|
||||
|
||||
| 过滤器 | 数据缺失时行为 |
|
||||
| --- | --- |
|
||||
| exclude_suspended / exclude_limit_locked | 素材必在（enriched 必算列）；缺失 = 数据本身异常 → fail 报错 |
|
||||
| exclude_st | instruments 名称缺失 → 过滤器跳过 + 降级注记（不静默假装过滤了） |
|
||||
| min_listing_days | listing_date 缺失的标的视为"不满足"剔除（保守），计数展示 |
|
||||
| cap_quantile_range | 历史股本缺失标的退出该过滤（不参与分位），降级注记 |
|
||||
| 退市股（未来） | 依赖新 dataset（§9）；无数据源时 universe 定义退化为"当前上市 ∪ 本地历史"，报告中永久注记幸存者偏差警示 |
|
||||
|
||||
---
|
||||
|
||||
## 5. 风险调整与基准（L-NEU）
|
||||
|
||||
### 5.1 NeutralizationSpec【设计】
|
||||
|
||||
```python
|
||||
@dataclass(frozen=True)
|
||||
class NeutralizationSpec:
|
||||
benchmark: str | None = "000001.SH" # 上证指数（index_const.py:12【现状】核心四只之一）；扩展指数见 §9
|
||||
return_basis: Literal["raw", "excess"] = "excess" # IC/分层收益口径
|
||||
method: Literal["none", "industry_demean", "industry_zscore", "regression_industry_size"] = "none"
|
||||
winsorize_sigma: float | None = 3.0 # 因子值截面截尾；None = 不截尾
|
||||
# neutralization_id = sha256(canonical_json)[:12]，进缓存键
|
||||
```
|
||||
|
||||
### 5.2 方法规格【设计】
|
||||
|
||||
- 超额收益：`r_ex = r_stock − r_bench`（基准同日收益；基准停市日沿用最近交易日，日历由数据轴驱动）。
|
||||
- `industry_demean`：`f' = f − mean_ind(f)`（THS 行业一级，ext preset 快照）。
|
||||
- `industry_zscore`：组内标准化 `f' = (f − μ_ind)/σ_ind`（σ=0 组 → null）。
|
||||
- `regression_industry_size`：`f ~ 1 + 行业哑变量 + log_float_mv` 的残差（逐日 OLS，Polars 表达式实现，n<30 或共线 → 回退 demean + 注记）。
|
||||
- 固定管线顺序：`宇宙过滤 → winsorize → 中性化 → 标准化(zscore 或 rank)`；顺序进 spec 哈希。
|
||||
- **行业快照局限**（当前归属回填历史）写入 `degraded` 注记并在报告显示；点时行业表到位后（§9）仅切换数据源，spec 不变。
|
||||
|
||||
### 5.3 报告口径并列【设计】
|
||||
|
||||
IC 报告同时输出三列：`原始 / 超额 / 超额+中性化`，默认排序以最后一列为准——旧结论可查，新结论更严，不静默替换。
|
||||
|
||||
---
|
||||
|
||||
## 6. 统计检验（L-INF）
|
||||
|
||||
**依赖原则**：后端当前无 scipy/statsmodels（已验证，§17），运行时保持零新增第三方依赖——NW/BH-FDR/DSR 全部以 numpy 手写实现（各约 20-40 行）；statsmodels 仅允许加入 uv dev 依赖组用于测试对拍，不进运行时 import。
|
||||
|
||||
### 6.1 统计量精确定义【设计】
|
||||
|
||||
| 统计量 | 定义 | 备注 |
|
||||
| --- | --- | --- |
|
||||
| Rank IC | 逐日 Spearman(factor_t, fwd_ret_{t→t+h})，现有口径不变（factor.py:740-750） | — |
|
||||
| IC t 值（朴素） | `t = mean(IC) / (std(IC, ddof=1)/√N)` | 仅作对照展示 |
|
||||
| IC t 值（NW） | Newey-West HAC 稳健标准误，滞后 `L = h`（h 日前瞻收益使 IC 序列存在 h−1 阶移动平均自相关） | **主口径**；numpy 手写 Bartlett 核加权，测试用固定黄金参考向量 + 可选 dev 组 statsmodels 对拍 |
|
||||
| ICIR | mean(IC)/std(IC)，已有 | — |
|
||||
| IC 自相关 & 半衰期 | ACF(1..10)；半衰期 = ACF 首次 < 0.5 的滞后（线性插值）；无收敛 → null | 换手率预期管理 |
|
||||
| 分层单调性 | Spearman(组序号, 组均超额收益) + 线性趋势斜率 t 值 | 判定"梯子是否成立" |
|
||||
| 多空 t 值 | 顶组−底组日超额收益序列的 NW t | 滞后 = 调仓周期的收益重叠阶数 |
|
||||
| BH-FDR q 值 | 对 optimizer/mining 排行榜全体 p 值（每行 = 其 OOS/IS 最优组合的 IC 或收益 t 值双尾 p）做 Benjamini-Hochberg，q_i = min_{j≥i}(N·p_j/j) 单调化 | 排行榜级，不进单因子报告 |
|
||||
| Deflated Sharpe | Bailey-López de Prado：以试验次数 N（挖掘 trial 预算已计数，mining.py:1041+）与偏度峰度校正 SR₀，DSR = Φ((SR−SR₀)·√(T−1) / √(1−γ̂₃SR+((γ̂₄−1)/4)SR²)) | 挖掘晋升报告展示"考虑搜索后的置信" |
|
||||
| 覆盖率/换手率 | 已有（factor.py:1123-1166），补充宇宙过滤后口径 | — |
|
||||
|
||||
### 6.2 显著性标注约定【设计】
|
||||
|
||||
|t| < 1.645 无标注；≥1.645 `*`(10%)；≥1.96 `**`(5%)；≥2.576 `***`(1%)。报告 UI 图标化，q ≥ 0.10 的挖掘候选禁止晋升（现有晋升门槛 mining.py:26-31 之上叠加，未达标给出具体差值）。
|
||||
|
||||
### 6.3 版本命名规则与报告完整字段【设计】
|
||||
|
||||
**方法论版本命名**（单一规则，全文档统一）：
|
||||
|
||||
- `factor_v3` = 本设计交付的因子研究方法论（三口径 + 统计检验 + 宇宙/中性化 spec 进键）；现有 `factor_v2`（factor.py:112【现状】）结果按旧版本读取展示，不重算。
|
||||
- `metrics_v2` = §7 指标统一后的口径；与 `factor_v3` 独立演进，报告分别携带。
|
||||
- 因子个体 `version`（FactorSpec）与研究方法论版本正交：因子公式变更不改方法论版本，反之亦然。
|
||||
|
||||
**IC 研究报告完整字段定义**（`POST /api/factor-research/ic` 响应，§10 示例为其节选）：
|
||||
|
||||
```text
|
||||
methodology_version: str # "factor_v3"
|
||||
factor_id / factor_version: str / int
|
||||
universe_id / neutralization_id: str # 两 spec 哈希
|
||||
universe: object # 回显生效 UniverseSpec
|
||||
neutralization: object # 回显生效 NeutralizationSpec
|
||||
date_range: {start, end, rebalance, n_groups}
|
||||
ic: {mean, std, icir, t_naive, t_newey_west, nw_lag, significance,
|
||||
half_life_days|null, acf: float[10], win_rate, coverage, n_days}
|
||||
ic_decay: [{horizon, ic_mean, icir}] # 沿用现有 1/3/5 日结构
|
||||
ic_yearly: [{year, ic_mean, icir, n_days}] # 沿用现有结构
|
||||
ic_by_basis: {raw, excess, excess_neutralized} × {mean, t_newey_west}
|
||||
monotonicity: {spearman, trend_t, verdict} # verdict ∈ 成立/弱/不成立
|
||||
groups: [{group, excess_return_annual, nav, turnover, n_names,
|
||||
t_stat, avg_name_count}] # 每组含 t 值
|
||||
long_short: {annual_return, t_newey_west, max_drawdown, executable_short: false}
|
||||
turnover_top_group: float
|
||||
costs: {commission_pct, stamp_tax_pct, slippage_bps, round_trip}
|
||||
universe_funnel: [{date, raw, after_suspended, after_limit_locked,
|
||||
after_st|null, after_new_listing|null, after_filters}]
|
||||
degraded: [str] # 如 st_not_point_in_time / industry_snapshot
|
||||
warnings: [str] # 非降级类提示（预热边界、覆盖不足等）
|
||||
```
|
||||
|
||||
新增字段全部带默认值，历史（factor_v2）缓存结果缺字段时前端显示为空，不报错。
|
||||
|
||||
---
|
||||
|
||||
## 7. 绩效指标统一（L-MET）
|
||||
|
||||
### 7.1 唯一口径【设计】
|
||||
|
||||
- **Sharpe**：净值曲线日收益 `r_t = nav_t/nav_{t−1} − 1`，`Sharpe = mean(r)/std(r, ddof=1) × √A`；无风险利率参数 `rf_annual`（默认 0，单位/年，日化按 A 折算）。引擎三种旧口径（逐笔/仅卖出日/净值）收敛为净值口径；前两者字段保留一个版本周期并标 `deprecated_mode`。
|
||||
- **年化天数 A**：默认 243（近五年 A 股实际均值区间），`metrics_methodology_version = "metrics_v2"`；报告展示口径徽章。
|
||||
- **年化收益**：`(nav_T/nav_0)^(A/n_bars) − 1`（统一按 K 线数折算，废除 365.25 自然日混用，engine.py:2871-2874 收敛）。
|
||||
- **MaxDD**：现有算法（峰值下限 1.0）不变。
|
||||
- **基准相对新增**：`excess_annual`、`tracking_error = std(r−r_b)×√A`、`information_ratio = mean(r−r_b)/std(r−r_b)×√A`、`beta/alpha`（OLS，rf 处理同上）、`excess_win_rate`。
|
||||
- `engine.py`/`strategy.py`/`factor.py` 全部改 import `app/factors/metrics.py`，禁止本地重算（Ruff 检查加入 noqa 禁用清单之外无豁免）。
|
||||
|
||||
---
|
||||
|
||||
## 8. 复合因子与策略接入（L-CMP）——核心章节
|
||||
|
||||
### 8.1 现有桥（【现状·已可用】，设计的锚点）
|
||||
|
||||
- 策略配置：`"scoring": {"factor_name": weight}` + `scoring_directions` 覆盖高低方向（`builtin/*.py` 均此形态）。
|
||||
- `scoring.py`：虚拟因子按需编译 Polars 表达式（`scoring_value_expr`）、依赖展开（`scoring_dependencies`）、预热推导（`scoring_warmup_bars`）。
|
||||
- 回测矩阵按 `score` 排序建仓（`engine.py` `max_positions` + `score_min/max`）。
|
||||
- 挖掘产物 = 因子排名组合（mining.py），候选库 `candidates.py` 已有 `factor`/`strategy` 双形态。
|
||||
|
||||
### 8.2 FactorCompositeSpec【设计】
|
||||
|
||||
```python
|
||||
@dataclass(frozen=True)
|
||||
class FactorCompositeSpec:
|
||||
id: str # ^cf_[a-z0-9_]{1,40}$；策略 scoring 里以 "cf_xxx" 引用
|
||||
version: int
|
||||
label: str
|
||||
factors: tuple[CompositeMember, ...] # 1..10 个成员
|
||||
transform: Literal["rank", "zscore"] = "rank" # 成员标准化方式（截面）
|
||||
weighting: Literal["manual", "equal", "icir", "max_ic"] = "manual"
|
||||
auto_weight_window: int = 504 # 自动权重的滚动窗口（交易日）
|
||||
direction: Literal["high", "low"] = "high"
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class CompositeMember:
|
||||
factor: str # 任意已注册因子 id（含 uf_/cf_ 前缀，禁止自引用，环检测）
|
||||
weight: float | None # manual 模式必填；自动模式忽略
|
||||
direction_override: Literal["high", "low"] | None = None
|
||||
```
|
||||
|
||||
**计算管线（顺序固定，进 spec 哈希）**：
|
||||
|
||||
```
|
||||
宇宙(可选, 默认不过滤以兼容现有策略)
|
||||
→ 各成员因子值（注册表展开，含 warmup 检查）
|
||||
→ winsorize(3σ, 可关)
|
||||
→ 截面 transform（rank/zscore）
|
||||
→ 方向统一（low → 取负）
|
||||
→ 加权求和（manual 权重归一化校验 |Σw−1|<1e-9；
|
||||
icir → w_i ∝ max(ICIR_i, 0)，ICIR 取 (t−auto_weight_window, t−1] 窗口——**权重只用于过去，严禁 t 日数据参与 t 日权重**；
|
||||
max_ic → 同窗口 mean(rank IC) 单调权重）
|
||||
→ 输出复合分 cf_xxx（rank 基础下近似 ∈ [−1,1]，文档声明分布性质）
|
||||
```
|
||||
|
||||
**接入策略（零引擎改动）**：
|
||||
|
||||
- 复合因子注册进注册表（kind="composite"），`scoring_value_expr` 机制天然支持：策略写 `"scoring": {"cf_hotmom": 0.6, "vol_ratio_5d": 0.2, "amount": 0.2}` 即生效；依赖/预热自动递归展开进矩阵构建，选股/回测/监控三端无需感知"这是复合因子"。
|
||||
- 挖掘产物一键导出为 CompositeSpec：mining 的排名组合本来 = factors+weights，导出即 `cf_mined_<run>`，闭合"挖掘 → 复合因子 → 策略"回路。
|
||||
- 前置校验：成员因子任一 warmup 超研究窗口 → 启动期注册成功但使用时返回明确"预热不足"错误（不产出半截分数）。
|
||||
|
||||
**策略侧引用形态（确切 JSON）**——复合因子编辑器"导出 scoring 片段"产出，直接粘贴进策略配置：
|
||||
|
||||
```json
|
||||
{
|
||||
"scoring": {"cf_hotmom": 0.6, "vol_ratio_5d": 0.2, "amount": 0.2},
|
||||
"scoring_directions": {"cf_hotmom": "high", "amount": "high"}
|
||||
}
|
||||
```
|
||||
|
||||
复合因子与普通字段混用、权重语义不变；`scoring_dependencies`/`scoring_warmup_bars` 自动递归展开（`scoring.py:91-104`【现状】机制不动，仅数据源换成注册表）。
|
||||
|
||||
### 8.3 一致性契约【设计】
|
||||
|
||||
同一 `cf_xxx` 在**因子研究（IC/分层）、选股、回测、监控**四端必须逐位同值——单测直接断言四路径对同一 (date,symbol) 的输出相等。这是 CONTRIBUTING 5.3"同一候选集和排序方向"的推广。
|
||||
|
||||
### 8.4 监控端数据流澄清【设计】
|
||||
|
||||
监控不重算复合因子：`monitor.py:1314-1322`【现状】消费的是**策略结果缓存**里的 `result.scores`（`score_min/score_max` 过滤）。因此复合因子进监控的路径 = 策略执行时算好分 → 结果缓存 → 监控读缓存。**实时行情线程零新增计算**（CONTRIBUTING 6.3 硬约束）。推论：修改复合因子定义后，必须走策略参数变更的既有失效链路（重算策略结果缓存 → 监控实例刷新），该链路已存在（CONTRIBUTING 5.1），设计只复用不新造。
|
||||
|
||||
### 8.5 自定义/复合因子的盘中行为【设计】
|
||||
|
||||
- 选股（盘后批量）：`incremental_safe=True` 的成员因子照常参与当日计算。
|
||||
- `incremental_safe=False` 成员（若有）：当日选股对该因子返回"预热/路径不足"的明确不可计算状态（对齐 CONTRIBUTING 5.1"空值不得伪装成零分"），UI 标注原因；**不降级用部分成员算半截复合分**。
|
||||
- 盘中增量路径（`pipeline.py:1795` `compute_enriched_today`【现状】）：自定义与复合因子默认不进入（§12 缓存策略），分时选股若引用则同样返回不可计算状态，盘后恢复。
|
||||
|
||||
---
|
||||
|
||||
## 9. 数据契约扩展（provider dataset）
|
||||
|
||||
新增 dataset 声明（capabilities.py 注册表 + 对应 provider 实现，均【设计】）：
|
||||
|
||||
| dataset | 内容 | 解锁能力 | 无数据源时 |
|
||||
| --- | --- | --- | --- |
|
||||
| `st_history` | 点时风险警示状态 | 历史涨跌停幅度修正、宇宙 ST 点时过滤 | 涨跌停用当前名推断 + 注记（现状） |
|
||||
| `delisted_kline` | 退市标的日 K + 退市维表 | 幸存者偏差修复（宇宙回补） | 报告永久幸存者注记 |
|
||||
| `industry_pit` | 点时行业归属 | 中性化升级为点时 | 行业快照 + 注记（现状） |
|
||||
| `index_ext` | 扩展指数日 K | 基准升级（当前限核心四只，index_const.py:12-15【现状】） | 基准限核心四只 |
|
||||
|
||||
各 dataset 完整字段 schema（provider 归一后落 Parquet，命名对齐现有 normalized 契约）：
|
||||
|
||||
**`st_history`**（分区 `data/parquet/st_history/`，按年）：
|
||||
|
||||
| 字段 | 类型 | 语义 |
|
||||
| --- | --- | --- |
|
||||
| symbol | str | 标准代码（如 600000.SH） |
|
||||
| flag_date | date | 状态生效日（戴帽/摘帽公告后的首个交易日） |
|
||||
| st_flag | bool | true=风险警示（ST/*ST），false=摘帽；行区间语义：自 flag_date 起至下一条记录 |
|
||||
| flag_type | str | "ST" / "*ST" / "摘帽"；缺失填 "ST" |
|
||||
| source | str | provider 标识 |
|
||||
|
||||
查询语义：`st_at(symbol, t) = flag_date ≤ t 的最后一条记录的 st_flag`（asof-backward）。同步：全量快照 + 增量 append，`(symbol, flag_date)` 去重幂等（对齐 kline_sync:358-367【现状】模式）。
|
||||
|
||||
**`delisted_kline`**：K 线部分复用 `daily` dataset 完整 schema（symbol/date/OHLC/volume/amount/…）；另需维表 `delisted_instruments`：
|
||||
|
||||
| 字段 | 类型 | 语义 |
|
||||
| --- | --- | --- |
|
||||
| symbol / name | str | 代码/退市前简称 |
|
||||
| list_date / delist_date | date | 上市/退市日 |
|
||||
| delist_reason | str | "面值"/"财务"/"重组"/"主动"/"其他"；缺失填 "其他" |
|
||||
|
||||
宇宙回补语义：`as_of=t 的可交易池 = instruments(上市≤t<退市) ∪ delisted(上市≤t<退市)`；`_resolve_universe`（daily_pipeline.py:92-125【现状】）扩展为两源合并。
|
||||
|
||||
**`industry_pit`**（分区 `data/parquet/industry_pit/`）：
|
||||
|
||||
| 字段 | 类型 | 语义 |
|
||||
| --- | --- | --- |
|
||||
| symbol | str | 标准代码 |
|
||||
| effective_date | date | 归属生效日 |
|
||||
| industry_l1 / industry_l2 | str | 一级/二级行业名（如 计算机/软件开发，对齐 market_mainline.py:35【现状】的两级口径） |
|
||||
| source | str | provider（如 ths） |
|
||||
|
||||
查询语义：asof-backward join（同 fundamentals.py:107-114【现状】模式）；effective_date 缺失 = 供应商不提供历史，整表降级为快照并触发 §5 注记。
|
||||
|
||||
**`index_ext`**：schema 与 `kline_index_daily` 完全一致（symbol/date/OHLC/volume/amount），仅标的白名单扩展（默认建议：000300.SH 沪深300、000905.SH 中证500、000852.SH 中证1000、000985.SH 中证全指）；白名单由 preset 配置声明，不进代码硬编码（对齐 ext_presets 模式）。
|
||||
|
||||
**能力注册示例**（provider 侧 `plugin.yaml` datasets 声明，对齐 `docs/plugin-development.md` 契约）：
|
||||
|
||||
```yaml
|
||||
datasets:
|
||||
st_history:
|
||||
enabled: true
|
||||
description: 点时风险警示状态（戴帽/摘帽区间）
|
||||
delisted_kline:
|
||||
enabled: true
|
||||
description: 退市标的日K + 退市维表
|
||||
industry_pit:
|
||||
enabled: false # 供应商无历史归属时声明 false，不注册能力
|
||||
description: 点时行业归属
|
||||
```
|
||||
|
||||
能力矩阵（`capabilities.py` 注册表）同步各 dataset 的展示元数据与路由偏好字段；provider 未声明 = 该能力全局不可用，研究路径按 §4.3 降级矩阵处理，不静默。
|
||||
|
||||
---
|
||||
|
||||
## 10. API 契约【设计】
|
||||
|
||||
新路由前缀 `/api/factors`（薄层，重计算在 services/factors_research.py 编排层）：
|
||||
|
||||
| 端点 | 方法 | 请求要点 | 响应要点 |
|
||||
| --- | --- | --- | --- |
|
||||
| `/api/factors/catalog` | GET | asset_type, group, stability 过滤 | 因子清单（含 formula_text/warmup/direction/pit/scale_free/usage_count） |
|
||||
| `/api/factors/validate` | POST | formula 文本 | 编译错误（位置+原因）或成功（依赖/预热推导） |
|
||||
| `/api/factors/preview` | POST | formula + symbols + date_range | 试算表格（最新 5 日 × 前 20 标的）+ 非空率 |
|
||||
| `/api/factors/custom` | GET/POST/DELETE | §3.4 JSON | CRUD；删除带引用清单 |
|
||||
| `/api/factor-research/ic` | POST | factor_id, universe_id/UniverseSpec, NeutralizationSpec, start/end, rebalance, n_groups | §6 全套统计 + 三口径并列 + 漏斗 filter_stats + degraded 注记 |
|
||||
| `/api/factor-research/composite` | GET/POST/DELETE | CompositeSpec | CRUD + 一键"作为评分字段试策略"跳转链接 |
|
||||
| `/api/factor-research/universes` | GET/POST | UniverseSpec 存档 | 命名宇宙 CRUD（研究配置复用） |
|
||||
|
||||
全部响应新增字段带默认值；错误响应含 `code/message/detail`，不泄漏内部栈（CONTRIBUTING 第 8 节）。SSE 进度复用现有回测 SSE 模式（长任务：批量 IC 扫描）。
|
||||
|
||||
**鉴权**：新路由经 `api/routes.py`【现状】注册，继承应用级部署口令鉴权（`docs/deploy-password.md` 模式），不引入独立权限模型。写操作（自定义因子/复合因子/宇宙存档 CRUD）只落 `data/user_data/`，路径校验沿用策略目录的防穿越规则（CONTRIBUTING 5.1 删除策略 fail-closed 要求同样适用）。
|
||||
|
||||
**并发**：批量 IC 扫描与复合分批量计算走 `services/heavy_job_limiter.py`【现状】限流，SSE 进度事件结构复用回测现有契约；用户取消走现有回测 worker 取消机制。
|
||||
|
||||
**核心端点示例**（其余端点按同构风格推导）：
|
||||
|
||||
`GET /api/factors/catalog?group=动量&asset_type=stock` →
|
||||
|
||||
```json
|
||||
{
|
||||
"factors": [
|
||||
{
|
||||
"id": "momentum_20d", "version": 1, "label": "20日动量", "group": "动量",
|
||||
"kind": "base", "formula_text": "20个交易日累计收益率",
|
||||
"direction": "high", "unit": "ratio", "warmup_bars": 20,
|
||||
"pit": false, "scale_free": true, "stability": "stable",
|
||||
"tags": ["momentum"], "usage_count": 7, "custom": false
|
||||
}
|
||||
],
|
||||
"total": 61, "degraded": []
|
||||
}
|
||||
```
|
||||
|
||||
`POST /api/factors/validate` `{"formula": "rank(ts_delta(close, -5))"}` →
|
||||
|
||||
```json
|
||||
{"ok": false, "errors": [{"code": "E005", "message": "负 shift: ts_delay 的 n 必须 ≥ 0（负数即未来函数）", "position": {"offset": 18, "line": 1}, "detail": {"n": -5}}]}
|
||||
```
|
||||
|
||||
`POST /api/factor-research/ic`：
|
||||
|
||||
```json
|
||||
{
|
||||
"factor_id": "momentum_20d",
|
||||
"universe": {"exclude_suspended": true, "exclude_limit_locked": true, "exclude_st": true, "min_listing_days": 60},
|
||||
"neutralization": {"benchmark": "000001.SH", "return_basis": "excess", "method": "industry_demean", "winsorize_sigma": 3.0},
|
||||
"start": "2023-01-01", "end": "2025-12-31",
|
||||
"rebalance": "monthly", "n_groups": 5
|
||||
}
|
||||
```
|
||||
|
||||
响应（节选，完整字段见 §6）：
|
||||
|
||||
```json
|
||||
{
|
||||
"methodology_version": "factor_v3",
|
||||
"universe_id": "a1b2c3d4e5f6", "neutralization_id": "9f8e7d6c5b4a",
|
||||
"ic": {"mean": 0.031, "icir": 0.42, "t_naive": 2.9, "t_newey_west": 1.87, "nw_lag": 1,
|
||||
"significance": "*", "half_life_days": 4, "acf": [0.21, 0.08, ...], "coverage": 0.97},
|
||||
"ic_by_basis": {"raw": {"mean": 0.041, "t_newey_west": 2.2}, "excess": {"mean": 0.031, "t_newey_west": 1.87},
|
||||
"excess_neutralized": {"mean": 0.019, "t_newey_west": 1.02}},
|
||||
"monotonicity": {"spearman": 0.9, "trend_t": 2.4, "verdict": "成立"},
|
||||
"universe_funnel": [{"date": "2025-12-31", "raw": 5412, "after_suspended": 5390, "after_limit_locked": 5320, "after_st": 5180, "after_new_listing": 5090}],
|
||||
"degraded": ["st_not_point_in_time", "industry_snapshot"],
|
||||
"turnover_top_group": 0.31, "costs_round_trip": 0.0013
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 11. 前端界面【设计】
|
||||
|
||||
按此前结论：**不新增顶层页面**，组件级落点。每个组件给出区块级线框与交互流：
|
||||
|
||||
1. **ResearchProfile 共享面板**（新组件，因子回测/挖掘/验证三视图共用）：UniverseSpec + NeutralizationSpec 的受控表单，可存档命名（对应 `/api/factor-research/universes`）；degraded 注记黄条；查询键含两 spec 哈希（queryKeys.ts 集中新增 `factorResearch` 键族）。
|
||||
线框：`[存档下拉 ▾] [另存为] | 折叠区1·宇宙（6 个过滤器开关/输入 + 漏斗摘要行） | 折叠区2·调整（基准/口径/方法/截尾） | [重置] [应用到当前视图]`；spec 哈希变化即触发查询键切换。
|
||||
2. **因子目录对话框**（因子回测 tab 内，仿 ResearchCandidatesDialog）：分组树 + 搜索 + 公式/方向/预热/PIT 徽章/引用数；"研究此因子"按钮回填选择器。
|
||||
线框：`左侧分组树(带计数) | 右侧表格[因子/公式/方向/预热/PIT/引用] | 底部[研究此因子][加入复合候选]`；"加入复合候选"把因子暂存到复合编辑器的选择篮（跨组件轻状态，放 TanStack Query 缓存而非全局 store）。
|
||||
3. **IC 报告增强**：t 值列（NW 主口径，显著性星标）、三口径并列、单调性判定、IC 半衰期、宇宙漏斗、降级注记条。
|
||||
线框：IC 摘要卡新增 `t(NW)=2.31** 半衰期=4d 单调性=成立(ρ=0.9)` 一行；分层表头新增口径切换 tab（原始/超额/超额+中性化），切换不改数据只换列；宇宙漏斗为横向递减条形（5200→4980→4890，hover 显示过滤器名）。
|
||||
4. **自定义因子编辑器**（Settings 新面板"因子库"，与信号库并列）：公式输入 + 算子速查侧栏 + 实时校验 + 试算预览 + 版本列表 + 引用关系展示。
|
||||
线框：`左列: 版本列表(当前高亮)+元信息表单 | 中列: 公式输入框(等宽,校验错误行内红标+光标定位) + 算子速查(点击插入) | 右列: 试算预览表(最新5日×前20标的+非空率) [校验] [试算] [保存]`；保存按钮在校验+试算双绿前禁用。
|
||||
5. **复合因子编辑器**（同 Settings 面板内 tab）：成员表（因子搜索、权重、方向）、自动权重开关与窗口、管线预览图；"试用于策略"向导生成 scoring 片段。
|
||||
线框：`上: 成员表[因子搜索器|方向|权重|剔除] + weighting 单选 + 窗口输入 | 中: 成员相关性热力图(§11-8, >0.8 对红标提示去重) + 管线预览(过滤→截尾→中性化→标准化→加权) | 下: [导出 scoring 片段] [试用于策略]`。
|
||||
6. **因子相关性探索器**（复合编辑器内嵌 + 因子回测 tab 的独立对话框）：任选 2-10 个因子，展示区间内日均截面秩相关矩阵热力图。后端复用 `mining.py:446` `compute_rank_correlation`【现状】抽出的公共函数，不新建第二套计算。
|
||||
7. **挖掘工作台**：排行榜加 t/q 值列与 DSR；候选卡新增"导出为复合因子"。
|
||||
8. 全部新组件覆盖 加载/空/错误/禁用/无权限 五态（CONTRIBUTING 第 7 节）**并在 1280px 常用宽度与窄屏（≤768px）检查截断、遮挡、弹窗可操作性**；前端类型同步进 `lib/api.ts`；所有轮询/长任务按钮带进行中禁用态。
|
||||
|
||||
**查询键新增**（`queryKeys.ts` 集中定义，spec 哈希必须进键）：
|
||||
|
||||
```text
|
||||
factorCatalog({assetType, group}) // 目录
|
||||
factorValidate() // mutation，无需键
|
||||
factorCustomList() / factorCustomPreview({formulaHash}) // CRUD / 试算
|
||||
factorIcReport({factorId, factorVersion, universeId, neuId,
|
||||
methodology, rangeHash, rebalance, nGroups})
|
||||
factorCompositeList() / factorCompositeEval({cmpSpecHash, matrixGeneration})
|
||||
factorCorrelation({factorIds[], rangeHash})
|
||||
researchUniverses()
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 12. 缓存与性能【设计】
|
||||
|
||||
| 缓存 | 键 | 失效 |
|
||||
| --- | --- | --- |
|
||||
| 因子 IC 报告 | `fr:ic:{factor_id}:{v}:{universe_id}:{neu_id}:{methodology}:{range_hash}:{rebalance}:{n_groups}` | enriched generation 变更或键任一分量变 |
|
||||
| 复合因子定义 | `fr:cmp:{id}:{spec_hash}` | 定义编辑 |
|
||||
| 复合分值（研究期） | `fr:cmpv:{cmp_spec_hash}:{matrix_generation}` | 矩阵重建 |
|
||||
| DSL 编译产物 | 进程内 LRU（表达式文本 → Expr） | 进程重启 |
|
||||
| 宇宙快照 | `fr:uni:{universe_id}:{matrix_generation}` | 矩阵重建 |
|
||||
|
||||
约束：复合/自定义因子**默认不物化进 enriched parquet**（避免用户定义污染核心管道与增量路径）；只在研究/评分请求期计算并按上表缓存。`incremental_safe=False` 的因子盘中路径直接缺失而非降级计算（对齐 pipeline 增量路径现有行为）。中性化逐日截面计算全部 Polars 表达式化；统计层 O(N·G) 极小。禁止任何新增逻辑进入实时行情线程（CONTRIBUTING 6.3）。
|
||||
|
||||
**性能预算（实现验收线，超线必须先优化再合入）**：
|
||||
|
||||
| 操作 | 预算 | 基准场景 |
|
||||
| --- | --- | --- |
|
||||
| 单因子 IC 全报告（含三口径+t 值+分层） | ≤ 现有报告耗时 × 1.3 | 全 A 股 × 3 年日线（现有 `factor.py` 同窗基线，PR 里附前后数据，CONTRIBUTING 6.3） |
|
||||
| DSL 编译（含校验） | ≤ 5ms/表达式 | 深度 12、token 200 上限样例 |
|
||||
| 复合因子单期截面计算 | ≤ 成员因子独立计算耗时之和 × 1.2 | 10 成员 × 全 A 股 |
|
||||
| 宇宙过滤全期 | ≤ 全期 IC 计算的 10% | 同上基准 |
|
||||
| 因子目录接口 | ≤ 50ms | 全量 62+ 因子元数据 |
|
||||
| 相关性探索器 | ≤ 现有 mining 同规模秩相关耗时 × 1.1 | 10 因子 × 1 年 |
|
||||
|
||||
---
|
||||
|
||||
## 13. 测试矩阵（最低要求清单）【设计】
|
||||
|
||||
| 模块 | 必测 |
|
||||
| --- | --- |
|
||||
| 注册表重构 | 特征化快照（62+ 因子两条计算路径逐位一致）；重复 id/未增版本拒绝启动 |
|
||||
| DSL | 每算子黄金用例（含 null/除零/σ=0/全常数）；负 shift 编译失败；深度/窗口/白名单越界拒绝；与手写 Polars 等价性；注入样例（`__import__`、列名穿越）拒绝 |
|
||||
| 宇宙 | 每过滤器独立单测（构造含 ST/停牌/涨停锁死/次新/微额的合成面板）；漏斗计数；降级注记触发 |
|
||||
| 中性化 | 合成数据数值断言（demean/zscore/回归残差 vs statsmodels 对拍）；行业缺组回退；基准停市日 |
|
||||
| 统计 | NW t：黄金参考向量（离线计算硬编码期望值）+ dev 组 statsmodels 对拍（可选）；FDR：BH 已知 p 向量解析解；DSR：已构造解析例（对称正态收益 + 已知试验数）；单调性边界（平梯/倒梯） |
|
||||
| 指标 | Sharpe/年化/超额/IR 已知序列解析解；243 口径回归 |
|
||||
| 复合因子 | 权重归一；**自动权重无未来函数**（t 日权重不随 t 日数据变化——篡改 t 日数据断言权重不变）；四端同值断言（§8.3）；环引用拒绝 |
|
||||
| API | 成功/空数据/编译错误/预热不足/无权限 |
|
||||
| 缓存 | 键覆盖测试（改 spec 必换键）；generation 失效 |
|
||||
| 前端 | pnpm build + 五态检查 |
|
||||
|
||||
---
|
||||
|
||||
## 14. 实施路线图（PR 粒度，每 PR 独立可合）
|
||||
|
||||
| PR | 内容 | 依赖 | 主要文件 | 分级 |
|
||||
| --- | --- | --- | --- | --- |
|
||||
| 1 | stats 模块：IC t(NW)/单调性/半衰期 + IC 报告新字段 | 无 | 新 `app/factors/stats.py` + factor.py 增量 | L2 |
|
||||
| 2 | metrics 统一 + metrics_v2 版本化 | PR-1 | 新 metrics.py；engine/strategy/factor 改引用 | L3（热点，最小接线） |
|
||||
| 3 | 宇宙构建器 + 过滤器 + IC 接入 + 缓存键 | PR-1 | 新 universe.py + factor.py | L2 |
|
||||
| 4 | 基准超额 + 中性化 + 三口径并列 | PR-3 | 新 neutralize.py + factor.py | L2 |
|
||||
| 5 | FDR + DSR 进 optimizer/mining 排行榜与晋升门槛 | PR-1 | optimizer/mining 增量 | L3 |
|
||||
| 6 | 因子注册表重构（特征化测试先行） | 无（可与 1-5 并行） | 新 registry.py；factor.py/scoring.py/pipeline.py 收口 | L3 |
|
||||
| 7 | 复合因子 + 策略 scoring 桥 + 挖掘导出 | PR-6 | 新 composite.py；scoring.py 最小接线 | L2→L3 |
|
||||
| 8 | DSL 编译器 + 自定义因子 CRUD + Settings 因子库 UI | PR-6 | 新 dsl/ + api + settings 前端 | L2 |
|
||||
| 9 | 前端：ResearchProfile + 目录 + 报告增强 | PR-3/4 后端就绪 | 前端组件族 | 前端 |
|
||||
| 10+ | 数据契约：st_history / delisted / industry_pit / index_ext | provider 侧 | capabilities + 各 provider | L1+L2 |
|
||||
|
||||
每个 PR 按 CONTRIBUTING 第 10 节模板出描述（问题/根因/方案/兼容/性能/验证/界面证据/回滚）。
|
||||
|
||||
**配置白名单联动**：`backtest/candidates.py:28-49`【现状】的 `_CONFIG_FIELDS["factor"]` 是冻结字段集，PR-3/PR-4 必须同步扩展 `universe`、`neutralization` 两个配置字段（沿用 `_MINING_SOURCE_CONFIG_FIELDS` 的 frozenset 合并模式），否则保存候选会静默丢弃 spec——这是缓存一致性之外的第二个容易漏的接线点，测试须覆盖"保存→载入→spec 哈希不变"。
|
||||
|
||||
**各 PR 回滚要点**：PR-1/3/4/5 新增模块 + 增量字段，回滚 = revert 即可（旧缓存键不含新分量，自动回旧路径）；PR-2 指标统一保留 `metrics_methodology` 开关，回滚 = 切回 v1 计算分支并保留数据；PR-6 注册表重构通过特征化测试保证行为等价，回滚 = revert（无持久化迁移）；PR-7/8 用户数据（自定义因子/复合因子 JSON）为新增目录，回滚代码后文件残留但不再加载，重新部署即恢复——**不存在任何需要用户手动清数据的回滚**（CONTRIBUTING 第 12 节红线）。
|
||||
|
||||
**文档同步任务**（各 PR 内完成，不单开）：PR-1/3/4 更新 `docs/features.md` 因子回测章节；PR-5 更新 `docs/mining.md` 门槛说明；PR-7 更新 `docs/strategy.md` 评分字段说明与 `操作说明书.md`；PR-8 更新 `docs/custom-data-source.md` 无关则跳过；本设计文档在每个 PR 合入后把对应条目从【设计】改标【已实现】。
|
||||
|
||||
---
|
||||
|
||||
## 15. 兼容性影响与风险清单
|
||||
|
||||
| 变更 | 功能影响 | 结果口径影响 | 缓解 |
|
||||
| --- | --- | --- | --- |
|
||||
| PR-1/5 统计字段 | 无 | 无（纯新增） | 字段默认值 |
|
||||
| PR-2 指标统一 | 无 | **Sharpe/年化数字变化（有意）** | metrics_v2 版本徽章；旧字段一版周期弃用 |
|
||||
| PR-3/4 宇宙/中性化 | 无 | IC/分层数字变化（有意，通常回落） | 默认开关显式；三口径并列；缓存键含 spec |
|
||||
| PR-6 注册表 | 无 | 要求逐位一致 | 特征化测试是合入硬门槛 |
|
||||
| PR-7/8 复合/DSL | 无（纯新增能力） | 无 | 注入列模式，custom_signals 先例 |
|
||||
| 数据契约 | 无 | 退市股回补后回测数字变化（修复） | 独立 dataset，无源时明确降级注记 |
|
||||
|
||||
剩余风险：~~① ts_rank 的 Polars 原生可用性需实现期确认~~【已解决：§17 验证 polars 1.40.1 `Expr.rolling_rank` 存在且行为正确】；② 行业快照回填历史的偏差在点时表到位前无法消除（注记透明化）；③ 幸存者偏差的根本修复依赖数据源，代码侧已尽（注记 + 回补接口预留）；④ ST 非点时在 st_history dataset 到位前仅能注记；⑤ 统计函数运行时零新增第三方依赖（后端当前无 scipy/statsmodels，已验证），NW/BH-FDR/DSR 以 numpy 手写实现，statsmodels 仅允许加入 uv dev 依赖组做测试对拍，不进运行时。
|
||||
|
||||
---
|
||||
|
||||
## 16. 明确不做清单（YAGNI 边界）
|
||||
|
||||
以下能力**刻意不在本设计范围内**，防止范围蔓延（依据 `docs/secondary-development.md` 第 10 节：不为未来可能出现的需求预埋框架）。出现真实需求时再按需立项：
|
||||
|
||||
| 不做项 | 理由 |
|
||||
| --- | --- |
|
||||
| 因子市场/分享/导入导出社区 | 单用户自托管定位，无真实需求 |
|
||||
| 全 Barra 风格回归（Beta/动量/流动性/非线性市值等十因子） | 数据与维护成本高；industry+size lite 已覆盖主要混杂，收益边际低 |
|
||||
| 自动机器学习/遗传规划因子搜索 | 与现有 beam search + 嵌套样本外定位重叠，且加剧多重检验问题 |
|
||||
| 港美股/加密资产因子 | 数据源与交易规则（T+0/无涨跌停）完全是另一套引擎 |
|
||||
| Tick 级/高频因子 | 分钟数据集能力有限，且与现有日线研究框架口径不同 |
|
||||
| 因子值的实时盘中推送（SSE 逐笔更新） | 违反实时热路径约束；监控经由策略结果缓存已覆盖时效需求 |
|
||||
| 复合因子权重在线学习/逐日再优化 | 自动权重窗口已是点时滚动；更细粒度会显著推高换手且引入过拟合面 |
|
||||
| ARIMA/VAR 预测、协整与配对交易 | 本平台定位是**横截面因子研究**；时间序列预测与统计套利是另一条业务线（指数择时/配对），数据、引擎与交易规则均不同，混入即范围蔓延 |
|
||||
| GARCH 全族 / 卡尔曼滤波 / 时变 Beta 状态空间模型 | EWMA 条件波动已覆盖日频主要价值；逐 symbol 递归拟合与全向量化管线冲突，机构级边际收益不抵维护成本（单 GARCH(1,1) 为 §2.4 的 experimental 按需项，不在冲突内） |
|
||||
| HMM/马尔可夫区制检测 | 现有情绪周期 6 阶段（启发式）+ 分环境 IC（factor.py:813-903【现状】）已覆盖区制条件分析；统计区制模型列为未来探索项不进本期 |
|
||||
| 独立权限体系（多用户/角色） | 应用级部署口令已满足自托管场景 |
|
||||
|
||||
---
|
||||
|
||||
## 17. 验证附录（本设计的验证记录）
|
||||
|
||||
> 验证日期 2026-09-04，基准 main@2ce8b4b1，后端 polars 1.40.1。分三部分：代码引用逐条核对、技术可行性实测、内部一致性检查。结论：**全部引用属实或已修正，可行性风险清零或已有替代方案，一致性检查通过**。
|
||||
|
||||
### 17.1 代码引用核对（【现状】条目逐条对账）
|
||||
|
||||
| 引用 | 核对内容 | 结果 |
|
||||
| --- | --- | --- |
|
||||
| factor.py:36-109 | FACTOR_COLUMNS 61 因子目录 | ✓（全文读取） |
|
||||
| factor.py:111/112/122 | warmup 120 / factor_v2 / n_groups=5 | ✓ |
|
||||
| factor.py:740-750 / 752-778 / 780-811 / 813-903 | Rank IC / 分年 / 衰减 / 分环境 | ✓（函数定义与实现均在引用区间） |
|
||||
| factor.py:1059-1067 / 1095-1098 | 双边佣金+印花税+滑点成本 / 每调仓期扣减 | ✓ |
|
||||
| factor.py:1123-1166 / 1001-1035 / 1294-1295 | 换手率 / tie-aware 分层 / executable_short=False | ✓ |
|
||||
| factor.py:1184-1186 / 1274-1277 | 年化系数匹配调仓频率的注释与实现 | ✓（注释原文核实） |
|
||||
| mining.py:26-31 / 91-100 / 1041-1073 / 1503-1530 / 446 | 晋升门槛 / purge30+embargo5 / trial 预算 / 折构造 / 秩相关 | ✓ |
|
||||
| mining.py:492-493 | `pl.corr(..., method="spearman")` 可用性 | ✓（代码在用，即证 API 存在） |
|
||||
| engine.py:2871-2874 / 2899-2903 / 2993-3030 / 3116-3122 | 365.25 年化 / 逐笔 Sharpe（含"非严格正确"注释）/ 仅卖出日聚合 / 净值口径 | ✓（四种口径全部原文核实） |
|
||||
| engine.py:51 / 54-94 / 903-933 | matching 默认 close_t / 成本模型 / 涨跌停与停牌成交闸 | ✓（全文精读） |
|
||||
| pipeline.py:970-984 | filter_halt_days（函数头 973） | ✓ |
|
||||
| pipeline.py:1795 | compute_enriched_today 盘中增量入口 | ✓ **（修正：初稿误引 1738+，该行实为复权因子读取；已改）** |
|
||||
| matrix.py:1476-1495 | _write_tradable_matrix | ✓ |
|
||||
| repository.py:1488 | get_index_daily | ✓ |
|
||||
| api/data.py:765 | listing_date 已暴露未用于研究 | ✓ |
|
||||
| fundamentals.py:107-115 | join_asof backward + date>_announce 严格公告日后 | ✓ |
|
||||
| share_capital.py:54-56 | announce_date 优先、period_end 兜底 | ✓ |
|
||||
| price_limits.py:87-101 | numpy_limit_pct_vectors（当前名推断） | ✓ |
|
||||
| capabilities.py:44-45 | 复权口径一致性"不做路由耦合"注释 | ✓ |
|
||||
| index_const.py:12-15 | 核心四只代码（000001.SH/399001.SZ/399006.SZ/000680.SH） | ✓ |
|
||||
| daily_pipeline.py:92-125 | _resolve_universe（CN_Equity_A 当前池） | ✓ |
|
||||
| kline_sync.py:358-367 | (symbol, trade_date) 去重 keep=last 原子合并 | ✓ |
|
||||
| scoring.py:13-51 / 53-66 / 91-104 / 108+ | VIRTUAL 依赖 35 项 / 预热表 / 依赖展开与预热推导 / scoring_value_expr | ✓（全文读取，附录A 由其逐字推导） |
|
||||
| monitor.py:1314-1322 | score_min/max 消费 result.scores | ✓ |
|
||||
| strategy/builtin/*.py | "scoring": {字段: 权重} 配置形态 | ✓（8 个内置策略抽样） |
|
||||
| candidates.py | factor/strategy 双形态候选配置字段 | ✓ |
|
||||
| services/heavy_job_limiter.py、backtest/numba_runtime.py、services/ext_presets.py、services/market_mainline.py:35 | 模块存在性 / 行业两级口径 | ✓ |
|
||||
|
||||
### 17.2 技术可行性实测（`uv run python` 于 backend 环境）
|
||||
|
||||
| 项 | 实测结果 | 设计影响 |
|
||||
| --- | --- | --- |
|
||||
| polars 版本 | 1.40.1（pyproject pin >=1.0） | — |
|
||||
| `Expr.rolling_rank` | **存在**，递增序列 4 点窗输出 [null,null,null,4,4,4] 行为正确 | §15 风险①**解除**，ts_rank 用原生实现 |
|
||||
| `Expr.ewm_std` / `ewm_var` | 存在 | f_ewma_vol 直接可实现 |
|
||||
| `pl.rolling_corr`（顶层）/ `pl.rolling_cov` | 存在且可算出正确相关值（3 点窗样例 0.6547） | ts_corr/ts_cov 编译目标为顶层函数 |
|
||||
| `pl.rolling_corr(...).over("symbol")` 分组组合 | **实测通过**：A/B 两组各自窗口内相关（0.6547 / −1.0），无串组 | ts_corr 多标的面板场景确认可行（初稿未验证，本轮补测） |
|
||||
| `Expr.rolling_corr` | **不存在**（仅顶层函数） | 算子表已按顶层函数修正 |
|
||||
| `Expr.rolling_arg_max` | **不存在**；rolling_map 为 Python 回调 | ts_argmax/ts_argmin v1 移除（§3.1 已注） |
|
||||
| rolling_quantile/var/mean/std/max/min/sum、shift、pct_change、diff、pow、sign、clip、log、cum_prod | 全部存在 | 其余算子无阻碍 |
|
||||
| `pl.corr` spearman | mining.py:492 在用 | 无阻碍 |
|
||||
| numba | 已安装（numba_runtime 真实存在） | GARCH/矩阵核扩展路径成立 |
|
||||
| statsmodels / scipy | **均未安装** | §6 零依赖原则：运行时 numpy 手写 NW/BH-FDR/DSR，statsmodels 仅可进 dev 依赖组 |
|
||||
|
||||
### 17.3 内部一致性检查
|
||||
|
||||
1. 缓存键 ↔ Spec 字段：UniverseSpec 7 字段、NeutralizationSpec 5 字段（含 winsorize_sigma 与管线顺序）全部进入各自 canonical_json/哈希 → 键覆盖完整（§4.2/§5.2 ↔ §12）。✓
|
||||
2. API ↔ 前端组件：catalog↔目录对话框、validate/preview/custom↔因子编辑器、ic↔报告增强、composite↔复合编辑器、universes↔ResearchProfile，无孤立端点或无后端组件。✓
|
||||
3. 测试矩阵 ↔ 模块：§13 十行覆盖 §2-§12 全部新增模块，无模块缺测试项。✓
|
||||
4. 路线图 ↔ 章节：PR-1↔§6.1、PR-2↔§7、PR-3↔§4、PR-4↔§5、PR-5↔§6.1(FDR/DSR)、PR-6↔§2、PR-7↔§8、PR-8↔§3、PR-9↔§11、PR-10↔§9，全部章节有落点 PR。✓
|
||||
5. §2.4 GARCH(1,1) experimental 与 §16"GARCH 全族不做"边界一致（单一按需项 vs 全族排除，§16 已加注）。✓
|
||||
6. EBNF ↔ 算子表：if_else/比较/逻辑均以函数调用或中缀产生式覆盖；优先级链 or<and<cmp<add<mul<unary 明确无歧义。✓
|
||||
7. 错误码目录 ↔ §3.2 校验规则：六条编译规则 + 三条运行时规则（E013/E015/E016）全部有码可映射。✓
|
||||
8. 附录A 计数：virtual 35 + base 非财务 19 + 财务 7 = 61，与 FACTOR_COLUMNS 行数一致；方向"待标定"仅用于振荡/环境依赖因子，不虚构。✓
|
||||
9. 三口径并列（§5.3）↔ API 响应 ic_by_basis 三键（§10）↔ 前端口径切换 tab（§11-3）三处一致。✓
|
||||
10. §16 不做清单 8 项与 §0 设计目标无冲突。✓
|
||||
+24
-1
@@ -52,7 +52,9 @@
|
||||
|
||||
**组合管理**:最大持仓数 · 敞口控制 · 等权 / 自定义仓位。
|
||||
|
||||
输出净值曲线 · 夏普 · 最大回撤 · 胜率 · 交易明细。SSE 流式进度支持切页重连,不会丢失回测任务。
|
||||
输出净值曲线 · 夏普 · 最大回撤 · 胜率 · 盈亏比(平均盈利单/平均亏损单,≥1 红 <1 绿) · 蒙卡回撤(中位/95% 边界,按成交顺序重抽样估计回撤分布)。SSE 流式进度支持切页重连,不会丢失回测任务。
|
||||
|
||||
**因子归因**:策略评分(scoring)引用了因子的回测,结果区提供「因子归因」tab——对比盈利单与亏损单在入场信号日的因子均值(胜单均值/败单均值/差值/样本数),判断每个因子在本轮交易中贡献了正筛选力还是在拖后腿。数据来自入场候选快照,零额外计算;因子名显示「英文 id + 中文标签」。覆盖 `meta.scoring` 非空的策略(内置与自定义 polars 策略),矩阵单因子排名类策略暂不产出归因。
|
||||
|
||||
**分钟策略回测**:分钟级策略逐交易日回放信号、以分钟收盘价入场,输出分钟级成交明细,与日线回测同一套 T+1 / 费用约束。
|
||||
|
||||
@@ -66,6 +68,27 @@
|
||||
|
||||
---
|
||||
|
||||
## 🔬 因子平台(Factors)
|
||||
|
||||
`/factors` 一级页,检验 / 因子库 / 编辑器 / 组合四个 tab,与策略线双向联动。设计与阶段规划见 [factor-platform-plan.md](./factor-platform-plan.md)。
|
||||
|
||||
**因子编辑器**:写 DSL 公式(如 `rank(-ts_sum(change_pct, 5))`)→ 校验语法(错误点击定位到字符)→ 试算 40 日 IC → 保存草稿。辅助输入:全部 25 个算子按「时序/截面/工具」分组点击插入(悬停显示签名与参数约束)、79 个可用字段双语 chip 点击插入、「从模板开始」下拉含经典模板与**我的因子**(导入已保存公式作为起点,非初始公式替换前有确认)。保存为草稿态,完整检验通过后在因子库激活。
|
||||
|
||||
**检验与组合**:服务端 IC/IR、Newey-West t 值、BH-FDR q 值、分层收益与多空对比;组合 tab 管理多因子权重。自定义/复合因子与策略评分共用同一条物化路径。
|
||||
|
||||
**因子 ↔ 策略联动**(四条桥):
|
||||
|
||||
| 方向 | 入口 | 说明 |
|
||||
| :--- | :--- | :--- |
|
||||
| 因子 → 策略 | 因子库「生成策略」 | 一键生成单因子排名策略(matrix_native,entry/exit 评分阈值 + top_rank 参数),保存为自定义策略可直接回测 |
|
||||
| 策略 → 因子 | 策略触发器编辑的 Zap 按钮 | 从因子快速创建条件信号(阈值给建议值),自动挂接到当前策略的入场/卖出触发器 |
|
||||
| AI → 因子 | 自定义信号的 AI 生成 | 提示词包含全部因子分组与 id,可直接写「因子·动量: momentum_20d > 0.8」类条件 |
|
||||
| 回测 → 因子 | 策略回测「因子归因」tab | 见上文回测引擎章节 |
|
||||
|
||||
**版本与生命周期**:公式修改保存为新版本并回草稿态(需重新检验);状态机 active / watch / retired / draft。
|
||||
|
||||
---
|
||||
|
||||
## 📡 监控中心(Monitor)
|
||||
|
||||
统一规则引擎,一个页面管理**四类监控**:
|
||||
|
||||
@@ -22,6 +22,7 @@
|
||||
- 后端源码扩展注册：`backend/app/custom/<module>.py`，支持 FastAPI 路由、启动钩子和通知格式化器。
|
||||
- 当前前端插槽：`layout.navigation.extra`、`stock-preview.footer`、`watchlist.toolbar`。
|
||||
- 当前后端继承点：`NotificationFormatter`。
|
||||
- 因子平台与策略线的既有桥接（已实现，二开时直接复用、勿重复实现）：因子库一键生成单因子排名策略（`GenerateFactorStrategyDialog` → `custom_factor_*`）；策略触发器引用因子条件信号（`AddFactorSignalDialog`，`csg_f_*`）；自定义信号 AI 提示词含因子分组（`custom_signals_ai.py`）；策略回测因子归因（`strategy.py` 的 `factor_attribution`，覆盖 `meta.scoring` 非空的策略）。
|
||||
|
||||
尚未实现、只能在真实需求出现后增加的能力：
|
||||
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
{
|
||||
"name": "tickflow-stock-panel-frontend",
|
||||
"private": true,
|
||||
"version": "0.2.2",
|
||||
"version": "0.2.3",
|
||||
"type": "module",
|
||||
"scripts": {
|
||||
"dev": "vite",
|
||||
|
||||
@@ -25,7 +25,7 @@ import {
|
||||
Star,
|
||||
ScanSearch,
|
||||
History,
|
||||
Pickaxe,
|
||||
Sigma,
|
||||
FileText,
|
||||
Settings,
|
||||
DatabaseZap,
|
||||
@@ -40,6 +40,7 @@ import {
|
||||
Sparkles,
|
||||
Layers2,
|
||||
Layers3,
|
||||
Zap,
|
||||
Landmark,
|
||||
RadioTower,
|
||||
CheckCircle2,
|
||||
@@ -84,9 +85,8 @@ const nav = [
|
||||
{ to: '/', label: '看板', icon: LayoutDashboard },
|
||||
{ to: '/watchlist', label: '自选', icon: Star },
|
||||
{ to: '/screener', label: '策略', icon: ScanSearch },
|
||||
{ to: '/factors', label: '因子', icon: Sigma },
|
||||
{ to: '/backtest', label: '回测', icon: History },
|
||||
{ to: '/mining', label: '挖掘', icon: Pickaxe },
|
||||
{ to: '/lots', label: '持仓提醒', icon: Layers2 },
|
||||
{ to: '/stock-analysis', label: '个股分析', icon: TrendingUp },
|
||||
{ to: '/limit-ladder', label: '连板梯队', icon: Flame },
|
||||
{ to: '/concept-analysis', label: '概念分析', icon: Layers3 },
|
||||
@@ -95,6 +95,8 @@ const nav = [
|
||||
{ to: '/monitor', label: '监控中心', icon: RadioTower },
|
||||
{ to: '/regime', label: '市场环境', icon: Gauge },
|
||||
{ to: '/abnormal', label: '异动监控', icon: Siren },
|
||||
{ to: '/lots', label: '持仓提醒', icon: Layers2 },
|
||||
{ to: '/signals', label: '信号库', icon: Zap },
|
||||
{ to: '/review', label: '复盘', icon: BookOpenCheck },
|
||||
{ to: '/indices', label: '指数', icon: BarChart3 },
|
||||
{ to: '/data', label: '数据', icon: Database },
|
||||
|
||||
@@ -585,7 +585,7 @@ export function StockPreviewDialog({ symbol, name, onClose, triggerInfo, navList
|
||||
}`}
|
||||
>
|
||||
{parseInt(w, 10)}日{' '}
|
||||
<span className={info.value >= 0 ? 'text-bull' : 'text-bear'}>{fmtPct(info.value, 1)}</span>
|
||||
<span className={info.value >= 0 ? 'text-bull' : 'text-bear'}>{fmtPct(info.value)}</span>
|
||||
<span className="text-muted"> / ±{(info.threshold * 100).toFixed(0)}%</span>
|
||||
<span className="text-muted"> · 接近{(info.closeness * 100).toFixed(0)}%</span>
|
||||
</span>
|
||||
|
||||
@@ -1,8 +1,9 @@
|
||||
import { useState } from 'react'
|
||||
import { useNavigate } from 'react-router-dom'
|
||||
import { Plus, Settings2 } from 'lucide-react'
|
||||
import { Plus, Settings2, Zap } from 'lucide-react'
|
||||
import type { CustomSignal } from '@/lib/api'
|
||||
import { CustomSignalDialog } from './CustomSignalDialog'
|
||||
import { AddFactorSignalDialog } from '@/pages/backtest/AddFactorSignalDialog'
|
||||
|
||||
interface Props {
|
||||
kind: 'entry' | 'exit'
|
||||
@@ -15,19 +16,31 @@ interface Props {
|
||||
export function SignalTriggerActions({ kind, signals, onChange, buttonClassName, iconClassName }: Props) {
|
||||
const navigate = useNavigate()
|
||||
const [open, setOpen] = useState(false)
|
||||
const [factorOpen, setFactorOpen] = useState(false)
|
||||
|
||||
const accent = kind === 'entry' ? 'hover:text-accent hover:border-accent/40' : 'hover:text-warning hover:border-warning/40'
|
||||
const btnCls = buttonClassName ?? 'rounded-btn border border-border bg-base p-1 text-muted transition-colors cursor-pointer'
|
||||
const iconCls = iconClassName ?? 'h-3.5 w-3.5'
|
||||
|
||||
const handleSaved = (signal: CustomSignal) => {
|
||||
if (signal.kind !== kind && signal.kind !== 'both') return
|
||||
const signalId = `csg_${signal.id}`
|
||||
const enableSignal = (signalId: string, signalKind: 'entry' | 'exit' | 'both') => {
|
||||
if (signalKind !== kind && signalKind !== 'both') return
|
||||
onChange(signals.includes(signalId) ? signals : [...signals, signalId])
|
||||
}
|
||||
|
||||
const handleSaved = (signal: CustomSignal) => enableSignal(`csg_${signal.id}`, signal.kind)
|
||||
const handleFactorCreated = (signal: { id: string; kind: 'entry' | 'exit' | 'both' }) =>
|
||||
enableSignal(`csg_${signal.id}`, signal.kind)
|
||||
|
||||
return (
|
||||
<>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setFactorOpen(true)}
|
||||
title="从因子快速创建条件（阈值给建议值）"
|
||||
className={`${btnCls} hover:border-amber-400/40 hover:text-amber-400`}
|
||||
>
|
||||
<Zap className={iconCls} />
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setOpen(true)}
|
||||
@@ -38,7 +51,7 @@ export function SignalTriggerActions({ kind, signals, onChange, buttonClassName,
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => navigate('/settings?tab=signals&highlight=signals')}
|
||||
onClick={() => navigate('/signals?highlight=signals')}
|
||||
title="去信号库"
|
||||
className={`${btnCls} hover:border-amber-400/40 hover:text-amber-400`}
|
||||
>
|
||||
@@ -51,6 +64,13 @@ export function SignalTriggerActions({ kind, signals, onChange, buttonClassName,
|
||||
onClose={() => setOpen(false)}
|
||||
onSaved={handleSaved}
|
||||
/>
|
||||
{factorOpen && (
|
||||
<AddFactorSignalDialog
|
||||
defaultKind={kind}
|
||||
onCreated={handleFactorCreated}
|
||||
onClose={() => setFactorOpen(false)}
|
||||
/>
|
||||
)}
|
||||
</>
|
||||
)
|
||||
}
|
||||
|
||||
+174
-2
@@ -1072,6 +1072,56 @@ export interface FactorColumn {
|
||||
desc: string
|
||||
}
|
||||
|
||||
// ===== Factor Library (注册表, P1) =====
|
||||
export type FactorKind = 'base' | 'virtual' | 'composite' | 'custom'
|
||||
export type FactorStability = 'stable' | 'experimental' | 'deprecated'
|
||||
|
||||
export interface FactorLibraryItem {
|
||||
id: string
|
||||
label: string
|
||||
group: string
|
||||
kind: FactorKind
|
||||
version: number
|
||||
formula: string
|
||||
direction: 'high' | 'low' | 'none'
|
||||
unit: string
|
||||
warmup_bars: number
|
||||
pit: boolean
|
||||
asset_types: string[]
|
||||
stability: FactorStability
|
||||
scale_free: boolean
|
||||
dependencies: string[]
|
||||
}
|
||||
|
||||
export interface FactorDslError {
|
||||
code: string
|
||||
message: string
|
||||
position: { offset: number; line: number }
|
||||
detail?: Record<string, unknown>
|
||||
}
|
||||
|
||||
export interface FactorValidateResponse {
|
||||
ok: boolean
|
||||
errors: FactorDslError[]
|
||||
dependencies: string[]
|
||||
referenced_factors: string[]
|
||||
warmup_bars: number
|
||||
cross_sectional: boolean
|
||||
}
|
||||
|
||||
export interface FactorTrialResponse {
|
||||
ok: boolean
|
||||
n_dates: number
|
||||
null_ratio: number | null
|
||||
ic_mean: number | null
|
||||
ic_std: number | null
|
||||
ir: number | null
|
||||
ic_win_rate: number | null
|
||||
t_newey_west?: number | null
|
||||
ic_series: { date: string; ic: number; n_symbols: number }[]
|
||||
message?: string
|
||||
}
|
||||
|
||||
export interface GroupStat {
|
||||
group: number
|
||||
label: string
|
||||
@@ -1113,6 +1163,12 @@ export interface FactorBatchItem {
|
||||
n_dates: number
|
||||
elapsed_ms: number
|
||||
error: string | null
|
||||
// metrics_v2 (P3): NW HAC t 值与 BH-FDR q 值; 样本不足为 null (前端降级经验规则)
|
||||
t_naive?: number | null
|
||||
t_newey_west?: number | null
|
||||
nw_lag?: number | null
|
||||
p_value?: number | null
|
||||
q_value?: number | null
|
||||
}
|
||||
|
||||
export interface FactorBatchResult {
|
||||
@@ -1185,8 +1241,8 @@ export interface MiningRun {
|
||||
run_id: string
|
||||
signature: string
|
||||
status: MiningRunStatus
|
||||
request: MiningRequestV1
|
||||
source?: 'manual' | 'scheduled'
|
||||
request: MiningRequestV1 & { auto?: boolean; auto_screening?: AutoScreening }
|
||||
source?: 'manual' | 'scheduled' | 'auto'
|
||||
created_at: string
|
||||
updated_at: string
|
||||
started_at?: string | null
|
||||
@@ -1198,6 +1254,37 @@ export interface MiningRun {
|
||||
summary?: MiningResultSummary | null
|
||||
}
|
||||
|
||||
// 自动挖掘 L1 筛选摘要 (后端 app/services/auto_mining.py 产出结构)
|
||||
export interface AutoScreening {
|
||||
profile: MiningBudgetProfile
|
||||
gate: { min_abs_ic: number; min_abs_ir: number; min_abs_t: number; max_q: number }
|
||||
screen_window: { start: string; end: string }
|
||||
n_total: number
|
||||
n_qualified: number
|
||||
pool: string[]
|
||||
pool_truncated: boolean
|
||||
qualified: Array<{ factor_name: string; label: string; group: string; ic: number | null; ir: number | null; t: number | null; q: number | null; direction: 1 | -1 }>
|
||||
failed: Array<{ factor_name: string; label: string; group: string; ic: number | null; ir: number | null; t: number | null; q: number | null; reason: string }>
|
||||
reason_counts: Record<string, number>
|
||||
elapsed_ms: number
|
||||
}
|
||||
|
||||
export interface MiningAutoStartPayload {
|
||||
asset_type?: 'stock' | 'etf'
|
||||
start?: string | null
|
||||
end?: string | null
|
||||
budget_profile?: MiningBudgetProfile
|
||||
correlation_threshold?: number
|
||||
force?: boolean
|
||||
}
|
||||
|
||||
export interface MiningAutoStartResponse {
|
||||
started: boolean
|
||||
reason?: string
|
||||
run?: MiningRun
|
||||
screening?: AutoScreening
|
||||
}
|
||||
|
||||
export interface MiningResultSummary {
|
||||
factor_count: number
|
||||
selected_factor_count: number
|
||||
@@ -1395,6 +1482,12 @@ export interface StrategyBacktestResult {
|
||||
drawdown_curve: { date: string; value: number }[]
|
||||
benchmark_curve?: { date: string; value: number; close?: number; name?: string; symbol?: string }[]
|
||||
trades: StrategyBacktestTrade[]
|
||||
/** v1 因子归因: 入场信号日因子值快照 × 成交盈亏 (胜/败单均值对比); 无评分因子或分钟路径时为 null */
|
||||
factor_attribution?: {
|
||||
factors: { factor: string; win_mean: number | null; lose_mean: number | null; win_n: number; lose_n: number }[]
|
||||
n_win: number
|
||||
n_lose: number
|
||||
} | null
|
||||
per_symbol_stats: {
|
||||
symbol: string
|
||||
n_trades: number
|
||||
@@ -2437,6 +2530,79 @@ export const api = {
|
||||
factorColumns: () =>
|
||||
request<{ columns: FactorColumn[] }>('/api/backtest/factor/columns'),
|
||||
|
||||
factorLibrary: (assetType?: 'stock' | 'etf') =>
|
||||
request<{ factors: FactorLibraryItem[] }>(
|
||||
`/api/factors${assetType ? `?asset_type=${assetType}` : ''}`,
|
||||
),
|
||||
|
||||
factorValidate: (formula: string) =>
|
||||
request<FactorValidateResponse>('/api/factors/validate', {
|
||||
method: 'POST',
|
||||
body: JSON.stringify({ formula }),
|
||||
}),
|
||||
|
||||
factorTrial: (payload: { formula: string; asset_type?: 'stock' | 'etf'; days?: number }) =>
|
||||
request<FactorTrialResponse>('/api/factors/trial', {
|
||||
method: 'POST',
|
||||
body: JSON.stringify(payload),
|
||||
}),
|
||||
|
||||
factorCustomCreate: (payload: {
|
||||
id?: string
|
||||
label: string
|
||||
group?: string
|
||||
formula: string
|
||||
description?: string
|
||||
direction?: 'high' | 'low' | 'none'
|
||||
}) =>
|
||||
request<{ ok: boolean; id: string; version: number }>('/api/factors/custom', {
|
||||
method: 'POST',
|
||||
body: JSON.stringify(payload),
|
||||
}),
|
||||
|
||||
factorCustomUpdate: (factorId: string, payload: {
|
||||
label: string
|
||||
group?: string
|
||||
formula: string
|
||||
description?: string
|
||||
direction?: 'high' | 'low' | 'none'
|
||||
}) =>
|
||||
request<{ ok: boolean; id: string; version: number; status: string }>(
|
||||
`/api/factors/custom/${encodeURIComponent(factorId)}/update`,
|
||||
{ method: 'POST', body: JSON.stringify(payload) },
|
||||
),
|
||||
|
||||
factorCompositeCreate: (payload: {
|
||||
id?: string
|
||||
label: string
|
||||
group?: string
|
||||
members: Record<string, number>
|
||||
description?: string
|
||||
direction?: 'high' | 'low' | 'none'
|
||||
}) =>
|
||||
request<{ ok: boolean; id: string; version: number }>('/api/factors/composite', {
|
||||
method: 'POST',
|
||||
body: JSON.stringify(payload),
|
||||
}),
|
||||
|
||||
factorDelete: (factorId: string, force = false) =>
|
||||
request<{ ok: boolean; id: string; removed_references?: string[] }>(
|
||||
`/api/factors/custom/${encodeURIComponent(factorId)}${force ? '?force=true' : ''}`,
|
||||
{ method: 'DELETE', quiet: true },
|
||||
),
|
||||
|
||||
factorSetStatus: (factorId: string, status: 'draft' | 'active' | 'watch' | 'retired') =>
|
||||
request<{ ok: boolean; id: string; status: string }>(
|
||||
`/api/factors/custom/${encodeURIComponent(factorId)}/status`,
|
||||
{ method: 'POST', body: JSON.stringify({ status }) },
|
||||
),
|
||||
|
||||
factorSetGroup: (factorId: string, group: string) =>
|
||||
request<{ ok: boolean; id: string; group: string }>(
|
||||
`/api/factors/custom/${encodeURIComponent(factorId)}/group`,
|
||||
{ method: 'POST', body: JSON.stringify({ group }) },
|
||||
),
|
||||
|
||||
factorRun: (payload: {
|
||||
factor_name: string
|
||||
symbols?: string[] | null
|
||||
@@ -2494,6 +2660,12 @@ export const api = {
|
||||
miningRun: (runId: string) =>
|
||||
request<MiningRun>(`/api/backtest/mining/runs/${encodeURIComponent(runId)}`),
|
||||
|
||||
miningAutoStart: (payload: MiningAutoStartPayload) =>
|
||||
request<MiningAutoStartResponse>('/api/backtest/mining/auto', {
|
||||
method: 'POST',
|
||||
body: JSON.stringify(payload),
|
||||
}),
|
||||
|
||||
miningStart: (payload: MiningRequestV1) =>
|
||||
request<MiningRun>('/api/backtest/mining/runs', {
|
||||
method: 'POST',
|
||||
|
||||
@@ -272,6 +272,67 @@ export async function startMining(payload: Parameters<typeof api.miningStart>[0]
|
||||
}
|
||||
}
|
||||
|
||||
/** 自动挖掘: L1 全量筛选在后端同步跑 (数十秒到几分钟)。
|
||||
* 模块级提交即返回, 弹窗可立刻关闭; 筛选期间用 progress 呈现, run 创建后走 SSE/轮询。
|
||||
*/
|
||||
export async function startAutoMining(payload: Parameters<typeof api.miningAutoStart>[0]) {
|
||||
closeEvents()
|
||||
localStorage.removeItem(ACTIVE_RUN_KEY)
|
||||
const token = connectionToken
|
||||
update({
|
||||
runId: null,
|
||||
isPending: true,
|
||||
cancelling: false,
|
||||
reconnecting: false,
|
||||
run: null,
|
||||
progress: { phase: 'screening', label: '正在全量筛选因子（约 1-3 分钟）' },
|
||||
result: null,
|
||||
previousResult: current.result ?? current.previousResult,
|
||||
error: null,
|
||||
})
|
||||
try {
|
||||
const data = await api.miningAutoStart(payload)
|
||||
if (connectionToken !== token || current.runId !== null) return
|
||||
if (data.started && data.run) {
|
||||
const run = data.run
|
||||
localStorage.setItem(ACTIVE_RUN_KEY, run.run_id)
|
||||
update({
|
||||
runId: run.run_id,
|
||||
run,
|
||||
progress: run.progress ?? current.progress,
|
||||
isPending: !TERMINAL_STATES.has(run.status),
|
||||
})
|
||||
if (TERMINAL_STATES.has(run.status)) {
|
||||
await refreshTerminalRun(run.run_id, run.status, run, token)
|
||||
} else {
|
||||
connect(run.run_id)
|
||||
}
|
||||
return
|
||||
}
|
||||
// 无达标因子: 不创建任务, 以 task.error 呈现原因分布 (挖掘页红字展示)
|
||||
const screening = data.screening
|
||||
const topReasons = Object.entries(screening?.reason_counts ?? {})
|
||||
.sort((a, b) => b[1] - a[1])
|
||||
.slice(0, 3)
|
||||
.map(([reason, count]) => `${reason} ${count} 个`)
|
||||
.join('；')
|
||||
update({
|
||||
isPending: false,
|
||||
progress: null,
|
||||
error: `${screening?.n_total ?? 0} 个因子里 0 个达标，未启动挖掘。`
|
||||
+ (topReasons ? `主要失败原因：${topReasons}。` : '')
|
||||
+ '可改选「探索」档（门槛更低）或拉长日期区间后重试。',
|
||||
})
|
||||
} catch (error) {
|
||||
if (connectionToken !== token || current.runId !== null) return
|
||||
update({
|
||||
isPending: false,
|
||||
progress: null,
|
||||
error: String((error as Error).message || error),
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
export async function cancelMining() {
|
||||
if (!current.runId || !current.isPending || current.cancelling) return
|
||||
const runId = current.runId
|
||||
|
||||
@@ -53,6 +53,7 @@ export const QK = {
|
||||
// Backtest
|
||||
backtestStatus: ['backtest-status'] as const,
|
||||
factorColumns: ['backtest-factor-columns'] as const,
|
||||
factorLibrary: (assetType: string) => ['factors-library', assetType] as const,
|
||||
miningRuns: ['backtest-mining-runs'] as const,
|
||||
miningAvailability: (assetType: string, profile: string, start: string, end: string) =>
|
||||
['backtest-mining-availability', assetType, profile, start, end] as const,
|
||||
|
||||
@@ -659,8 +659,9 @@ function DeviationView({ onPreview }: {
|
||||
{ruleChips()}
|
||||
</div>
|
||||
<p className="mt-2.5 border-t border-border/60 pt-2 text-[10px] leading-relaxed text-muted">
|
||||
口径说明: 偏离值 = 个股 N 日累计涨跌幅 − 对应指数同期涨跌幅 (沪: 上证A指/上证指数,
|
||||
深: 深证A指/深证成指, 北: 北证50)。阈值为交易所异常波动披露标准的近似值, 仅供风险提示,
|
||||
口径说明: 偏离值 = 个股 N 日累计涨跌幅 − 对应指数同期涨跌幅 (沪主板: 上证A指/上证指数,
|
||||
科创板: 科创50, 深主板: 深证A指/深证成指, 创业板: 创业板综指, 北: 北证50)。
|
||||
阈值为交易所异常波动披露标准的近似值, 仅供风险提示,
|
||||
不构成监管认定。每只股票在 3日/10日/30日 三档各算一个接近度 (|偏离值| ÷ 该档阈值,
|
||||
阈值随板块不同; 2026-07-06 起主板风险警示股票与普通股票同口径), 表格「接近度」列与状态取三档中的最高值,
|
||||
来源窗口的偏离值颜色加重显示、其余窗口淡化; ≥100% 已触发、≥70% 边缘、≥50% 观察。
|
||||
|
||||
@@ -1,21 +1,15 @@
|
||||
import { useState } from 'react'
|
||||
import { Navigate, useSearchParams } from 'react-router-dom'
|
||||
import { BarChart3, BookmarkCheck, FlaskConical, ShieldCheck } from 'lucide-react'
|
||||
import { useEffect, useState } from 'react'
|
||||
import { Navigate, useLocation, useNavigate, useSearchParams } from 'react-router-dom'
|
||||
import { BookmarkCheck, FlaskConical, ShieldCheck } from 'lucide-react'
|
||||
import { PageHeader } from '@/components/PageHeader'
|
||||
import { FactorDiscovery } from './backtest/FactorDiscovery'
|
||||
import { ResearchCandidatesDialog } from './backtest/ResearchCandidatesDialog'
|
||||
import { RobustnessValidation } from './backtest/RobustnessValidation'
|
||||
import { StrategyBacktest } from './backtest/StrategyBacktest'
|
||||
import { type ResearchCandidate } from '@/lib/api'
|
||||
|
||||
type Tab = 'factor' | 'strategy' | 'robustness'
|
||||
type Tab = 'strategy' | 'robustness'
|
||||
|
||||
const MODES: Record<Tab, { title: string; subtitle: string; icon: typeof BarChart3 }> = {
|
||||
factor: {
|
||||
title: '因子',
|
||||
subtitle: '批量筛选与单因子检验',
|
||||
icon: BarChart3,
|
||||
},
|
||||
const MODES: Record<Tab, { title: string; subtitle: string; icon: typeof FlaskConical }> = {
|
||||
strategy: {
|
||||
title: '策略',
|
||||
subtitle: '现有策略评估与候选沉淀',
|
||||
@@ -30,17 +24,34 @@ const MODES: Record<Tab, { title: string; subtitle: string; icon: typeof BarChar
|
||||
|
||||
export function Backtest() {
|
||||
const [searchParams, setSearchParams] = useSearchParams()
|
||||
const location = useLocation()
|
||||
const navigate = useNavigate()
|
||||
const requestedTab = searchParams.get('tab')
|
||||
const [candidatesOpen, setCandidatesOpen] = useState(false)
|
||||
// 候选「载入复测」: 弹窗选定 → 关闭弹窗切到策略页 → StrategyBacktest 消费后清空
|
||||
const [pendingLoad, setPendingLoad] = useState<ResearchCandidate | null>(null)
|
||||
|
||||
// 旧链接兼容: 挖掘已升级为一级路由 /mining, 保留 run/candidate 参数重定向
|
||||
if (requestedTab === 'mining') {
|
||||
// 跨页「载入复测」: 因子页候选弹窗经 router state 传入 (location.state), 消费后清除防止刷新重复载入
|
||||
const stateCandidate = (location.state as { loadCandidate?: ResearchCandidate } | null)?.loadCandidate ?? null
|
||||
useEffect(() => {
|
||||
if (!stateCandidate) return
|
||||
setPendingLoad(stateCandidate)
|
||||
navigate({ pathname: location.pathname, search: location.search }, { replace: true })
|
||||
}, [stateCandidate, location.pathname, location.search, navigate])
|
||||
|
||||
// 旧链接兼容: 因子已升级为一级路由 /factors, 保留其余参数重定向
|
||||
if (requestedTab === 'factor') {
|
||||
const next = new URLSearchParams(searchParams)
|
||||
next.delete('tab')
|
||||
const search = next.toString()
|
||||
return <Navigate to={search ? `/mining?${search}` : '/mining'} replace />
|
||||
return <Navigate to={search ? `/factors?${search}` : '/factors'} replace />
|
||||
}
|
||||
|
||||
// 旧链接兼容: 挖掘已并入因子页 (/factors?tab=mining), 保留 run/candidate 参数重定向
|
||||
if (requestedTab === 'mining') {
|
||||
const next = new URLSearchParams(searchParams)
|
||||
next.set('tab', 'mining')
|
||||
return <Navigate to={`/factors?${next.toString()}`} replace />
|
||||
}
|
||||
|
||||
const activeTab: Tab = requestedTab && requestedTab in MODES
|
||||
@@ -101,7 +112,6 @@ export function Backtest() {
|
||||
/>
|
||||
|
||||
<main className="min-h-0 flex-1 px-3 pb-3 pt-3 lg:px-4 lg:pb-4">
|
||||
{activeTab === 'factor' && <FactorDiscovery />}
|
||||
{activeTab === 'strategy' && (
|
||||
<StrategyBacktest
|
||||
loadCandidate={pendingLoad}
|
||||
|
||||
@@ -0,0 +1,127 @@
|
||||
import { useState } from 'react'
|
||||
import { useNavigate, useSearchParams } from 'react-router-dom'
|
||||
import { BookmarkCheck, Combine, LibrarySquare, PenLine, Pickaxe, Sigma } from 'lucide-react'
|
||||
import { PageHeader } from '@/components/PageHeader'
|
||||
import { FactorDiscovery } from './backtest/FactorDiscovery'
|
||||
import { FactorComposite } from './factors/FactorComposite'
|
||||
import { FactorEditor } from './factors/FactorEditor'
|
||||
import { FactorLibrary } from './factors/FactorLibrary'
|
||||
import { MiningWorkbench } from './backtest/MiningWorkbench'
|
||||
import { ResearchCandidatesDialog } from './backtest/ResearchCandidatesDialog'
|
||||
|
||||
type Tab = 'inspect' | 'library' | 'editor' | 'composite' | 'mining'
|
||||
|
||||
const TABS: Record<Tab, { title: string; icon: typeof Sigma }> = {
|
||||
inspect: { title: '检验', icon: Sigma },
|
||||
library: { title: '因子库', icon: LibrarySquare },
|
||||
editor: { title: '编辑器', icon: PenLine },
|
||||
composite: { title: '组合', icon: Combine },
|
||||
mining: { title: '挖掘', icon: Pickaxe },
|
||||
}
|
||||
|
||||
export function Factors() {
|
||||
const [searchParams, setSearchParams] = useSearchParams()
|
||||
const [candidatesOpen, setCandidatesOpen] = useState(false)
|
||||
const navigate = useNavigate()
|
||||
|
||||
const requestedTab = searchParams.get('tab')
|
||||
const activeTab: Tab = requestedTab === 'library' || requestedTab === 'editor' || requestedTab === 'composite' || requestedTab === 'mining'
|
||||
? requestedTab
|
||||
: 'inspect'
|
||||
const focusFactor = activeTab === 'inspect' ? searchParams.get('focus') ?? '' : ''
|
||||
|
||||
const changeTab = (tab: Tab) => {
|
||||
const next = new URLSearchParams(searchParams)
|
||||
next.set('tab', tab)
|
||||
next.delete('focus') // 切 tab 清除单因子聚焦
|
||||
next.delete('edit') // 切 tab 清除编辑器载入目标
|
||||
setSearchParams(next, { replace: true })
|
||||
}
|
||||
|
||||
return (
|
||||
<div className="flex min-h-full flex-col bg-base xl:h-full">
|
||||
<PageHeader
|
||||
title="因子"
|
||||
subtitle={<span className="hidden md:inline">{activeTab === 'library' ? '因子注册表目录' : activeTab === 'mining' ? '嵌套样本外因子与策略挖掘' : '因子研究与检验'}</span>}
|
||||
className="shrink-0 flex-wrap gap-x-4 gap-y-2 bg-base/95 px-3 lg:flex-nowrap lg:px-5"
|
||||
right={(
|
||||
<div className="flex w-full min-w-0 items-center gap-1.5 sm:gap-2 lg:w-auto">
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setCandidatesOpen(true)}
|
||||
aria-label="打开候选方案"
|
||||
title="候选方案"
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 rounded-btn border border-border bg-surface px-2 text-[11px] font-medium text-secondary transition-colors hover:border-accent/40 hover:text-accent sm:px-2.5 sm:text-xs"
|
||||
>
|
||||
<BookmarkCheck className="h-3.5 w-3.5" />
|
||||
<span>候选方案</span>
|
||||
</button>
|
||||
<span className="h-5 w-px shrink-0 bg-border" aria-hidden="true" />
|
||||
<nav className="min-w-0 flex-1 overflow-x-auto lg:flex-none" aria-label="因子视图">
|
||||
<div className="inline-flex min-w-max items-center gap-0.5 rounded-btn border border-border bg-surface/80 p-0.5">
|
||||
{(Object.keys(TABS) as Tab[]).map(tab => {
|
||||
const mode = TABS[tab]
|
||||
const Icon = mode.icon
|
||||
const active = activeTab === tab
|
||||
return (
|
||||
<button
|
||||
key={tab}
|
||||
type="button"
|
||||
onClick={() => changeTab(tab)}
|
||||
aria-current={active ? 'page' : undefined}
|
||||
className={`inline-flex h-7 items-center gap-1 rounded-[5px] px-1.5 text-[11px] font-medium transition-colors sm:gap-1.5 sm:px-2.5 sm:text-xs ${active
|
||||
? 'bg-accent text-white shadow-sm'
|
||||
: 'text-secondary hover:bg-elevated hover:text-foreground'
|
||||
}`}
|
||||
>
|
||||
<Icon className="hidden h-3.5 w-3.5 sm:block" />
|
||||
{mode.title}
|
||||
</button>
|
||||
)
|
||||
})}
|
||||
</div>
|
||||
</nav>
|
||||
</div>
|
||||
)}
|
||||
/>
|
||||
|
||||
<main className="min-h-0 flex-1 px-3 pb-3 pt-3 lg:px-4 lg:pb-4 xl:overflow-hidden">
|
||||
{activeTab === 'inspect' && <FactorDiscovery focusFactor={focusFactor} />}
|
||||
{activeTab === 'editor' && <FactorEditor key={searchParams.get('edit') ?? ''} editId={searchParams.get('edit') ?? ''} />}
|
||||
{activeTab === 'composite' && <FactorComposite />}
|
||||
{activeTab === 'mining' && (
|
||||
<div className="h-full min-h-0 xl:overflow-y-auto">
|
||||
<MiningWorkbench />
|
||||
</div>
|
||||
)}
|
||||
{activeTab === 'library' && (
|
||||
<FactorLibrary
|
||||
onInspect={factorId => {
|
||||
const next = new URLSearchParams(searchParams)
|
||||
next.set('tab', 'inspect')
|
||||
next.set('focus', factorId)
|
||||
setSearchParams(next)
|
||||
}}
|
||||
onEdit={factorId => {
|
||||
const next = new URLSearchParams(searchParams)
|
||||
next.set('tab', 'editor')
|
||||
next.set('edit', factorId)
|
||||
setSearchParams(next)
|
||||
}}
|
||||
/>
|
||||
)}
|
||||
</main>
|
||||
|
||||
{candidatesOpen && (
|
||||
<ResearchCandidatesDialog
|
||||
onClose={() => setCandidatesOpen(false)}
|
||||
onLoadStrategy={candidate => {
|
||||
setCandidatesOpen(false)
|
||||
// 跨页「载入复测」: 候选经 router state 交给回测策略页消费
|
||||
navigate('/backtest?tab=strategy', { state: { loadCandidate: candidate } })
|
||||
}}
|
||||
/>
|
||||
)}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
@@ -1,37 +0,0 @@
|
||||
import { useState } from 'react'
|
||||
import { BookmarkCheck } from 'lucide-react'
|
||||
import { PageHeader } from '@/components/PageHeader'
|
||||
import { MiningWorkbench } from './backtest/MiningWorkbench'
|
||||
import { ResearchCandidatesDialog } from './backtest/ResearchCandidatesDialog'
|
||||
|
||||
export function Mining() {
|
||||
const [candidatesOpen, setCandidatesOpen] = useState(false)
|
||||
|
||||
return (
|
||||
<div className="flex min-h-full flex-col bg-base">
|
||||
<PageHeader
|
||||
title="挖掘"
|
||||
subtitle={<span className="hidden md:inline">嵌套样本外因子与策略挖掘</span>}
|
||||
className="shrink-0 flex-wrap gap-x-4 gap-y-2 bg-base/95 px-3 lg:flex-nowrap lg:px-5"
|
||||
right={(
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setCandidatesOpen(true)}
|
||||
aria-label="打开候选方案"
|
||||
title="候选方案"
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 rounded-btn border border-border bg-surface px-2 text-[11px] font-medium text-secondary transition-colors hover:border-accent/40 hover:text-accent sm:px-2.5 sm:text-xs"
|
||||
>
|
||||
<BookmarkCheck className="h-3.5 w-3.5" />
|
||||
<span>候选方案</span>
|
||||
</button>
|
||||
)}
|
||||
/>
|
||||
|
||||
<main className="min-h-0 flex-1 px-3 pb-3 pt-3 lg:px-4 lg:pb-4">
|
||||
<MiningWorkbench />
|
||||
</main>
|
||||
|
||||
{candidatesOpen && <ResearchCandidatesDialog onClose={() => setCandidatesOpen(false)} />}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
@@ -6,14 +6,13 @@
|
||||
import { useState } from 'react'
|
||||
import { useSearchParams } from 'react-router-dom'
|
||||
import { motion } from 'framer-motion'
|
||||
import { BarChart3, Database, Radio, SlidersHorizontal, Sparkles, Settings2, Zap, PanelLeftClose, PanelLeftOpen, Clock3 } from 'lucide-react'
|
||||
import { BarChart3, Database, Radio, SlidersHorizontal, Sparkles, Settings2, PanelLeftClose, PanelLeftOpen, Clock3 } from 'lucide-react'
|
||||
import { SettingsAIPanel } from './settings/AI'
|
||||
import { SettingsMonitoringPanel } from './settings/Monitoring'
|
||||
import { SettingsExtPagesPanel } from './settings/ExtPages'
|
||||
import { SettingsMenuSettingsPanel } from './settings/MenuSettings'
|
||||
import { SettingsTimeoutPanel } from './settings/Timeout'
|
||||
import { SettingsSystemPanel } from './settings/System'
|
||||
import { SettingsCustomSignalsPanel } from './settings/CustomSignals'
|
||||
import { SettingsDataSourcesPanel } from './settings/DataSources'
|
||||
import { PageHeader } from '@/components/PageHeader'
|
||||
import { cn } from '@/lib/cn'
|
||||
@@ -35,7 +34,6 @@ const TABS: readonly TabDef[] = [
|
||||
{ key: 'ai', label: 'AI 设置', icon: Sparkles, panel: SettingsAIPanel },
|
||||
{ key: 'monitoring', label: '实时监控', icon: Radio, panel: SettingsMonitoringPanel },
|
||||
{ key: 'ext-pages', label: '扩展页面', icon: BarChart3, panel: SettingsExtPagesPanel },
|
||||
{ key: 'signals', label: '信号库', icon: Zap, panel: SettingsCustomSignalsPanel },
|
||||
{ key: 'timeout', label: '网络设置', icon: Clock3, panel: SettingsTimeoutPanel },
|
||||
{ key: 'menus', label: '菜单设置', icon: SlidersHorizontal, panel: SettingsMenuSettingsPanel },
|
||||
{ key: 'system', label: '系统设置', icon: Settings2, panel: SettingsSystemPanel },
|
||||
|
||||
@@ -1,11 +1,13 @@
|
||||
import { useEffect, useRef, useState } from 'react'
|
||||
import { useSearchParams } from 'react-router-dom'
|
||||
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
|
||||
import { Plus, Trash2, Zap, Settings2, Lock } from 'lucide-react'
|
||||
import { Lock, Plus, Settings2, Trash2, Zap } from 'lucide-react'
|
||||
import { api, type CustomSignal } from '@/lib/api'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
import { BUILTIN_SIGNAL_DEFINITIONS, type SignalKind } from '@/lib/signals'
|
||||
import { CustomSignalDialog } from '@/components/signals/CustomSignalDialog'
|
||||
import { Skeleton } from '@/components/data/Skeleton'
|
||||
import { PageHeader } from '@/components/PageHeader'
|
||||
import { AnchorWrap } from '@/lib/useCardFlash'
|
||||
|
||||
type SignalSection = 'builtin' | 'custom'
|
||||
@@ -17,12 +19,29 @@ const KIND_CLASS: Record<SignalKind, string> = {
|
||||
both: 'bg-muted/10 text-muted',
|
||||
}
|
||||
|
||||
export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string } = {}) {
|
||||
/** 信号库独立页: 内置只读信号 + 自定义条件信号 (csg_*), 策略/回测/监控统一取用。 */
|
||||
export function Signals() {
|
||||
const [searchParams] = useSearchParams()
|
||||
const highlight = searchParams.get('highlight') ?? ''
|
||||
|
||||
return (
|
||||
<div className="flex flex-col h-full">
|
||||
<PageHeader title="信号库" subtitle="内置预计算信号与自定义条件信号, 供策略 / 回测 / 监控统一使用" />
|
||||
<div className="min-h-0 flex-1 overflow-y-auto px-5 py-4">
|
||||
<div className="mx-auto max-w-6xl">
|
||||
<SignalsBody highlight={highlight} />
|
||||
</div>
|
||||
</div>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
function SignalsBody({ highlight }: { highlight: string }) {
|
||||
const qc = useQueryClient()
|
||||
const list = useQuery({ queryKey: QK.customSignals, queryFn: api.customSignalsList })
|
||||
const options = useQuery({ queryKey: QK.customSignalsOptions, queryFn: api.customSignalsOptions })
|
||||
|
||||
const [activeSection, setActiveSection] = useState<SignalSection>('builtin')
|
||||
const [activeSection, setActiveSection] = useState<SignalSection>('custom')
|
||||
const [showForm, setShowForm] = useState(false)
|
||||
const [editing, setEditing] = useState<CustomSignal | null>(null)
|
||||
const [confirmingDeleteId, setConfirmingDeleteId] = useState<string | null>(null)
|
||||
@@ -31,10 +50,6 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
const fields = options.data?.fields ?? []
|
||||
const signals = list.data?.signals ?? []
|
||||
const enabledCustomSignals = signals.filter(sig => sig.enabled).length
|
||||
const tabs = [
|
||||
{ key: 'builtin' as const, label: '内置信号', count: BUILTIN_SIGNAL_DEFINITIONS.length, hint: '系统提供，只读' },
|
||||
{ key: 'custom' as const, label: '自定义信号', count: signals.length, hint: `${enabledCustomSignals} 个已启用` },
|
||||
]
|
||||
|
||||
useEffect(() => () => {
|
||||
if (resetDeleteTimer.current) clearTimeout(resetDeleteTimer.current)
|
||||
@@ -86,35 +101,16 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
resetDeleteTimer.current = setTimeout(() => setConfirmingDeleteId(null), 3000)
|
||||
}
|
||||
|
||||
const tabs = [
|
||||
{ key: 'custom' as const, label: '自定义信号', count: signals.length, hint: `${enabledCustomSignals} 个已启用` },
|
||||
{ key: 'builtin' as const, label: '内置信号', count: BUILTIN_SIGNAL_DEFINITIONS.length, hint: '系统预计算，只读' },
|
||||
]
|
||||
|
||||
return (
|
||||
<div className="max-w-6xl space-y-6">
|
||||
<AnchorWrap highlight={highlight} anchor="signals">
|
||||
<section className="rounded-2xl border border-border bg-surface p-6 bg-[radial-gradient(circle_at_top_right,rgba(234,179,8,0.12),transparent_38%)]">
|
||||
<div className="flex flex-col gap-4 md:flex-row md:items-start md:justify-between">
|
||||
<div>
|
||||
<div className="text-[11px] uppercase tracking-[0.2em] text-amber-400/80">信号库</div>
|
||||
<h2 className="mt-2 text-2xl font-semibold tracking-tight text-foreground">统一查看策略、回测与监控可用信号</h2>
|
||||
<p className="mt-2 max-w-3xl text-sm leading-6 text-secondary">
|
||||
内置信号由系统预计算，作为只读信号库展示；自定义信号可用「字段 + 运算符 + 值」组合条件创建，保存后可在策略、回测与监控中选择使用。
|
||||
</p>
|
||||
</div>
|
||||
<button
|
||||
onClick={openNew}
|
||||
className="inline-flex items-center justify-center gap-1.5 rounded-btn bg-amber-500/90 px-3 py-1.5 text-xs font-medium text-base hover:bg-amber-500 transition-colors"
|
||||
>
|
||||
<Plus className="h-3.5 w-3.5" />
|
||||
新建自定义信号
|
||||
</button>
|
||||
</div>
|
||||
|
||||
<div className="mt-5 grid grid-cols-1 gap-3 md:grid-cols-3">
|
||||
<StatCard label="内置信号" value={BUILTIN_SIGNAL_DEFINITIONS.length} hint="系统提供，只读" />
|
||||
<StatCard label="自定义信号" value={signals.length} hint="用户创建，可编辑" />
|
||||
<StatCard label="已启用自定义" value={enabledCustomSignals} hint="会注入 csg_* 列" />
|
||||
</div>
|
||||
|
||||
<div className="mt-5 rounded-card border border-border bg-base/60 p-1.5">
|
||||
<div className="grid grid-cols-1 gap-1.5 md:grid-cols-2">
|
||||
<div className="space-y-4">
|
||||
{/* 紧凑工具条: 分段切换 + 新建 */}
|
||||
<div className="flex flex-wrap items-center justify-between gap-2">
|
||||
<div className="inline-flex rounded-btn border border-border bg-surface/80 p-0.5">
|
||||
{tabs.map(tab => {
|
||||
const active = activeSection === tab.key
|
||||
return (
|
||||
@@ -122,95 +118,53 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
key={tab.key}
|
||||
type="button"
|
||||
onClick={() => setActiveSection(tab.key)}
|
||||
className={`rounded-btn px-4 py-3 text-left transition-colors ${active ? 'bg-amber-500/15 text-amber-300 shadow-sm' : 'text-secondary hover:bg-elevated hover:text-foreground'}`}
|
||||
aria-current={active ? 'page' : undefined}
|
||||
className={`inline-flex h-7 items-center gap-1.5 rounded-[5px] px-2.5 text-[11px] font-medium transition-colors sm:text-xs ${active
|
||||
? 'bg-amber-500/90 text-white shadow-sm'
|
||||
: 'text-secondary hover:bg-elevated hover:text-foreground'}`}
|
||||
>
|
||||
<div className="flex items-center justify-between gap-3">
|
||||
<span className="text-sm font-medium">{tab.label}</span>
|
||||
<span className={`rounded px-2 py-0.5 text-[11px] ${active ? 'bg-amber-400/15 text-amber-300' : 'bg-elevated text-muted'}`}>{tab.count}</span>
|
||||
</div>
|
||||
<div className="mt-1 text-[11px] text-muted">{tab.hint}</div>
|
||||
{tab.label}
|
||||
<span className={`rounded px-1 py-px text-[10px] font-mono ${active ? 'bg-black/15 text-white' : 'bg-elevated text-muted'}`}>
|
||||
{tab.count}
|
||||
</span>
|
||||
</button>
|
||||
)
|
||||
})}
|
||||
</div>
|
||||
</div>
|
||||
</section>
|
||||
</AnchorWrap>
|
||||
|
||||
{activeSection === 'builtin' && (
|
||||
<section className="rounded-card border border-border bg-surface p-5 space-y-4">
|
||||
<div className="flex flex-col gap-2 md:flex-row md:items-end md:justify-between">
|
||||
<div>
|
||||
<div className="flex items-center gap-2">
|
||||
<Lock className="h-3.5 w-3.5 text-muted" />
|
||||
<h3 className="text-sm font-medium text-foreground">内置信号</h3>
|
||||
<span className="rounded bg-elevated px-1.5 py-0.5 text-[10px] text-muted">只读</span>
|
||||
</div>
|
||||
<p className="mt-1 text-xs text-muted">这些信号由系统在 enriched 数据中预计算，策略选择器会直接展示。</p>
|
||||
</div>
|
||||
<div className="text-[11px] text-muted">ID 前缀：<span className="font-mono text-foreground/70">signal_</span></div>
|
||||
</div>
|
||||
<div className="grid grid-cols-1 gap-3 md:grid-cols-2 xl:grid-cols-3">
|
||||
{BUILTIN_SIGNAL_DEFINITIONS.map(sig => (
|
||||
<div key={sig.id} className="rounded-card border border-border bg-base p-4">
|
||||
<div className="flex items-start justify-between gap-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<h4 className="text-sm font-medium text-foreground truncate">{sig.name}</h4>
|
||||
<span className={`rounded px-1.5 py-0.5 text-[10px] ${KIND_CLASS[sig.kind]}`}>
|
||||
{KIND_LABEL[sig.kind]}
|
||||
</span>
|
||||
</div>
|
||||
<p className="mt-1 text-[11px] text-muted font-mono truncate">{sig.id}</p>
|
||||
</div>
|
||||
<span className="shrink-0 rounded border border-border bg-elevated px-1.5 py-0.5 text-[10px] text-muted">{sig.category}</span>
|
||||
</div>
|
||||
<p className="mt-3 text-xs leading-5 text-secondary">{sig.description}</p>
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
</section>
|
||||
)}
|
||||
|
||||
{activeSection === 'custom' && (
|
||||
<section className="rounded-card border border-border bg-surface p-5 space-y-4">
|
||||
<div className="flex flex-col gap-3 md:flex-row md:items-center md:justify-between">
|
||||
<div>
|
||||
<div className="flex items-center gap-2">
|
||||
<Zap className="h-3.5 w-3.5 text-amber-400" />
|
||||
<h3 className="text-sm font-medium text-foreground">自定义信号</h3>
|
||||
<span className="rounded bg-amber-400/10 px-1.5 py-0.5 text-[10px] text-amber-400">可配置</span>
|
||||
</div>
|
||||
<p className="mt-1 text-xs text-muted">这些信号由你定义，可启用/停用，并在策略、回测与监控中作为 csg_* 信号使用。</p>
|
||||
</div>
|
||||
<div className="flex items-center gap-3">
|
||||
<span className="text-[11px] text-muted">自定义信号保存为 <span className="font-mono text-secondary">csg_*</span> 列，条件字段可用行情指标与全部注册因子</span>
|
||||
<button
|
||||
onClick={openNew}
|
||||
className="inline-flex items-center justify-center gap-1.5 rounded-btn border border-amber-400/30 bg-amber-400/5 px-3 py-1.5 text-xs font-medium text-amber-400 hover:bg-amber-400/10 transition-colors"
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 rounded-btn border border-amber-400/30 bg-amber-400/5 px-3 text-xs font-medium text-amber-400 transition-colors hover:bg-amber-400/10"
|
||||
>
|
||||
<Plus className="h-3.5 w-3.5" />
|
||||
新建自定义信号
|
||||
新建信号
|
||||
</button>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
<div className="grid grid-cols-1 md:grid-cols-2 gap-4">
|
||||
{activeSection === 'custom' && (
|
||||
<AnchorWrap highlight={highlight} anchor="signals">
|
||||
<section className="rounded-card border border-border bg-surface p-4">
|
||||
<div className="grid grid-cols-1 gap-3 md:grid-cols-2">
|
||||
{signals.map(sig => (
|
||||
<div key={sig.id} className="rounded-card border border-border bg-base p-4">
|
||||
<div className="flex items-start justify-between gap-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<h3 className="text-sm font-medium text-foreground truncate">{sig.name}</h3>
|
||||
<h3 className="truncate text-sm font-medium text-foreground">{sig.name}</h3>
|
||||
<span className={`rounded px-1.5 py-0.5 text-[10px] ${KIND_CLASS[sig.kind]}`}>
|
||||
{KIND_LABEL[sig.kind]}
|
||||
</span>
|
||||
{!sig.enabled && <span className="rounded bg-muted/10 px-1.5 py-0.5 text-[10px] text-muted">已停用</span>}
|
||||
</div>
|
||||
<p className="mt-1 text-[11px] text-muted font-mono truncate">csg_{sig.id}</p>
|
||||
<p className="mt-1 truncate font-mono text-[11px] text-muted">csg_{sig.id}</p>
|
||||
</div>
|
||||
<div className="flex items-center gap-1 shrink-0">
|
||||
<button onClick={() => toggleEnabled(sig)} title={sig.enabled ? '停用' : '启用'} className={`p-1 rounded cursor-pointer ${sig.enabled ? 'text-emerald-400 hover:bg-emerald-400/10' : 'text-muted hover:bg-elevated'}`}>
|
||||
<div className="flex shrink-0 items-center gap-1">
|
||||
<button onClick={() => toggleEnabled(sig)} title={sig.enabled ? '停用' : '启用'} className={`cursor-pointer rounded p-1 ${sig.enabled ? 'text-emerald-400 hover:bg-emerald-400/10' : 'text-muted hover:bg-elevated'}`}>
|
||||
<Zap className="h-3.5 w-3.5" />
|
||||
</button>
|
||||
<button onClick={() => openEdit(sig)} className="p-1 rounded text-muted hover:text-accent hover:bg-accent/10 cursor-pointer" title="编辑">
|
||||
<button onClick={() => openEdit(sig)} className="cursor-pointer rounded p-1 text-muted hover:bg-accent/10 hover:text-accent" title="编辑">
|
||||
<Settings2 className="h-3.5 w-3.5" />
|
||||
</button>
|
||||
{confirmingDeleteId === sig.id ? (
|
||||
@@ -218,7 +172,7 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
onClick={() => handleDeleteClick(sig)}
|
||||
disabled={del.isPending}
|
||||
title="再次点击确认删除"
|
||||
className="inline-flex items-center gap-1 rounded-md bg-danger/15 px-1.5 py-0.5 text-[10px] font-medium text-danger border border-danger/30 animate-pulse cursor-pointer disabled:opacity-50"
|
||||
className="inline-flex animate-pulse cursor-pointer items-center gap-1 rounded-md border border-danger/30 bg-danger/15 px-1.5 py-0.5 text-[10px] font-medium text-danger disabled:opacity-50"
|
||||
>
|
||||
<Trash2 className="h-2.5 w-2.5" />确认
|
||||
</button>
|
||||
@@ -226,7 +180,7 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
<button
|
||||
onClick={() => handleDeleteClick(sig)}
|
||||
disabled={del.isPending}
|
||||
className="p-1 rounded text-muted hover:text-danger hover:bg-danger/10 cursor-pointer disabled:opacity-50"
|
||||
className="cursor-pointer rounded p-1 text-muted hover:bg-danger/10 hover:text-danger disabled:opacity-50"
|
||||
title="删除"
|
||||
>
|
||||
<Trash2 className="h-3.5 w-3.5" />
|
||||
@@ -237,7 +191,7 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
<div className="mt-3 space-y-1">
|
||||
{sig.conditions.map((c, i) => (
|
||||
<div key={i} className="flex items-center gap-1.5 text-[11px] text-secondary">
|
||||
<span className="text-muted/50 w-6 text-right">{i === 0 ? '当' : '且'}</span>
|
||||
<span className="w-6 text-right text-muted/50">{i === 0 ? '当' : '且'}</span>
|
||||
<span className="font-mono text-foreground/80">{fieldWithDays(c.left, c.leftDays, fields)}</span>
|
||||
<span className="font-mono text-muted">{c.op}</span>
|
||||
<span className="font-mono text-foreground/80">
|
||||
@@ -252,19 +206,48 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
))}
|
||||
{list.isLoading &&
|
||||
Array.from({ length: 2 }).map((_, i) => (
|
||||
<div key={`sk-${i}`} className="rounded-card border border-border bg-base p-4 space-y-3">
|
||||
<div key={`sk-${i}`} className="space-y-3 rounded-card border border-border bg-base p-4">
|
||||
<Skeleton w="w-1/2" h="h-4" />
|
||||
<Skeleton w="w-1/3" h="h-3" />
|
||||
<Skeleton h="h-4" />
|
||||
</div>
|
||||
))}
|
||||
{!list.isLoading && signals.length === 0 && (
|
||||
<div className="rounded-card border border-border bg-base px-5 py-10 text-center text-sm text-muted md:col-span-2">
|
||||
暂无自定义信号，点击右上角「新建自定义信号」。
|
||||
<div className="rounded-card border border-dashed border-border px-5 py-10 text-center text-sm text-muted md:col-span-2">
|
||||
暂无自定义信号。可用「字段 + 运算符 + 值」组合条件创建，或从检验页因子行一键生成；也可让 AI 按描述生成。
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
</section>
|
||||
</AnchorWrap>
|
||||
)}
|
||||
|
||||
{activeSection === 'builtin' && (
|
||||
<section className="rounded-card border border-border bg-surface p-4">
|
||||
<div className="mb-3 flex items-center gap-2 text-xs text-muted">
|
||||
<Lock className="h-3.5 w-3.5" />
|
||||
系统在 enriched 数据中预计算，策略选择器直接展示，只读。
|
||||
</div>
|
||||
<div className="grid grid-cols-1 gap-3 md:grid-cols-2 xl:grid-cols-3">
|
||||
{BUILTIN_SIGNAL_DEFINITIONS.map(sig => (
|
||||
<div key={sig.id} className="rounded-card border border-border bg-base p-4">
|
||||
<div className="flex items-start justify-between gap-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<h4 className="truncate text-sm font-medium text-foreground">{sig.name}</h4>
|
||||
<span className={`rounded px-1.5 py-0.5 text-[10px] ${KIND_CLASS[sig.kind]}`}>
|
||||
{KIND_LABEL[sig.kind]}
|
||||
</span>
|
||||
</div>
|
||||
<p className="mt-1 truncate font-mono text-[11px] text-muted">{sig.id}</p>
|
||||
</div>
|
||||
<span className="shrink-0 rounded border border-border bg-elevated px-1.5 py-0.5 text-[10px] text-muted">{sig.category}</span>
|
||||
</div>
|
||||
<p className="mt-3 text-xs leading-5 text-secondary">{sig.description}</p>
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
</section>
|
||||
)}
|
||||
|
||||
<CustomSignalDialog open={showForm} signal={editing} onClose={closeForm} />
|
||||
@@ -272,16 +255,6 @@ export function SettingsCustomSignalsPanel({ highlight }: { highlight?: string }
|
||||
)
|
||||
}
|
||||
|
||||
function StatCard({ label, value, hint }: { label: string; value: number; hint: string }) {
|
||||
return (
|
||||
<div className="rounded-card border border-border/80 bg-base/70 px-4 py-3">
|
||||
<div className="text-[11px] text-muted">{label}</div>
|
||||
<div className="mt-1 text-2xl font-semibold text-foreground">{value}</div>
|
||||
<div className="mt-0.5 text-[11px] text-muted">{hint}</div>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
function fieldLabel(key: string, fields: { key: string; label: string }[]): string {
|
||||
return fields.find(f => f.key === key)?.label ?? key
|
||||
}
|
||||
@@ -0,0 +1,196 @@
|
||||
import { useState } from 'react'
|
||||
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
|
||||
import { Zap } from 'lucide-react'
|
||||
import { Modal } from '@/components/Modal'
|
||||
import { toast } from '@/components/Toast'
|
||||
import { api, type FactorBatchItem } from '@/lib/api'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
|
||||
const INPUT_CLS = 'h-8 w-full rounded-input border border-border bg-surface px-2.5 text-xs text-foreground focus:border-accent focus:outline-none'
|
||||
|
||||
/** 按因子族给默认阈值 (RSI/KDJ 经典区间, z 分 ±2, 有界位置 0.2/0.8, 乖离 ±5%)。 */
|
||||
function suggestThreshold(factorId: string, op: '>' | '<'): number {
|
||||
if (/^rsi_/.test(factorId)) return op === '<' ? 30 : 70
|
||||
if (/^kdj_/.test(factorId)) return op === '<' ? 20 : 80
|
||||
if (factorId.includes('_z_')) return op === '<' ? -2 : 2
|
||||
if (/(boll_position|close_position|position_240d)/.test(factorId)) return op === '<' ? 0.2 : 0.8
|
||||
if (/_bias/.test(factorId)) return op === '<' ? -0.05 : 0.05
|
||||
return 0
|
||||
}
|
||||
|
||||
interface CreatedSignal {
|
||||
id: string
|
||||
kind: 'entry' | 'exit' | 'both'
|
||||
name: string
|
||||
}
|
||||
|
||||
/**
|
||||
* 因子 → 条件信号快捷创建。
|
||||
* - 检验页传入 item: 固定因子, 方向按 IC 符号预填。
|
||||
* - 触发器场景不传 item: 显示因子选择器 (注册表全量), 阈值按因子族给建议值;
|
||||
* onCreated 回调让调用方把新信号立即启用为触发器。
|
||||
*/
|
||||
export function AddFactorSignalDialog({
|
||||
item, defaultKind = 'entry', onCreated, onClose,
|
||||
}: {
|
||||
item?: FactorBatchItem
|
||||
defaultKind?: 'entry' | 'exit'
|
||||
onCreated?: (signal: CreatedSignal) => void
|
||||
onClose: () => void
|
||||
}) {
|
||||
const queryClient = useQueryClient()
|
||||
const [selectedId, setSelectedId] = useState('')
|
||||
const library = useQuery({
|
||||
queryKey: QK.factorLibrary('all'),
|
||||
queryFn: () => api.factorLibrary(),
|
||||
enabled: !item,
|
||||
staleTime: 60_000,
|
||||
})
|
||||
const libItem = !item ? library.data?.factors.find(f => f.id === selectedId) : undefined
|
||||
const selected = item
|
||||
? { factor_name: item.factor_name, label: item.label, group: item.group }
|
||||
: libItem
|
||||
? { factor_name: libItem.id, label: libItem.label, group: libItem.group }
|
||||
: undefined
|
||||
|
||||
const icPositive = (item?.ic_mean ?? 0) >= 0
|
||||
const [op, setOp] = useState<'>' | '<'>(item ? (icPositive ? '>' : '<') : '>')
|
||||
const [threshold, setThreshold] = useState(
|
||||
item ? String(suggestThreshold(item.factor_name, icPositive ? '>' : '<')) : '',
|
||||
)
|
||||
const [name, setName] = useState(item ? `${item.label}·${icPositive ? '高值利多' : '低值利多'}` : '')
|
||||
const [kind, setKind] = useState<'entry' | 'exit' | 'both'>(defaultKind)
|
||||
const thresholdValid = threshold.trim() !== '' && Number.isFinite(Number(threshold))
|
||||
|
||||
const pickFactor = (factorId: string) => {
|
||||
setSelectedId(factorId)
|
||||
const factor = library.data?.factors.find(f => f.id === factorId)
|
||||
if (!factor) return
|
||||
setThreshold(String(suggestThreshold(factorId, op)))
|
||||
setName(`${factor.label}·${op === '>' ? '高值' : '低值'}`)
|
||||
}
|
||||
const pickOp = (nextOp: '>' | '<') => {
|
||||
setOp(nextOp)
|
||||
if (!item && selected) setThreshold(String(suggestThreshold(selected.factor_name, nextOp)))
|
||||
}
|
||||
|
||||
const signalId = selected
|
||||
? `f_${selected.factor_name}`.replace(/[^a-z0-9_]/g, '').slice(0, 40)
|
||||
: ''
|
||||
|
||||
const save = useMutation({
|
||||
mutationFn: () => api.customSignalSave({
|
||||
id: signalId,
|
||||
name: name.trim() || selected!.label,
|
||||
kind,
|
||||
enabled: true,
|
||||
conditions: [{ left: selected!.factor_name, op, right: threshold.trim(), leftDays: 0, rightDays: 0 }],
|
||||
}),
|
||||
onSuccess: () => {
|
||||
onCreated?.({ id: signalId, kind, name: name.trim() || selected!.label })
|
||||
toast('已保存为自定义信号（信号库可查看编辑）', 'success')
|
||||
void queryClient.invalidateQueries({ queryKey: QK.customSignals })
|
||||
onClose()
|
||||
},
|
||||
onError: (error: Error) => toast(`保存失败 · ${error.message}`, 'error'),
|
||||
})
|
||||
|
||||
const grouped = !item && library.data
|
||||
? Object.entries(library.data.factors.reduce<Record<string, typeof library.data.factors>>((acc, f) => {
|
||||
;(acc[f.group] ??= []).push(f)
|
||||
return acc
|
||||
}, {})).sort(([a], [b]) => a.localeCompare(b, 'zh'))
|
||||
: []
|
||||
|
||||
return (
|
||||
<Modal
|
||||
onClose={onClose}
|
||||
labelledBy="add-factor-signal-title"
|
||||
panelClassName="flex max-h-[86vh] w-[92vw] max-w-md flex-col overflow-hidden border border-border bg-surface shadow-xl rounded-card"
|
||||
>
|
||||
<div className="flex items-start justify-between gap-3 border-b border-border px-4 py-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<Zap className="h-4 w-4 shrink-0 text-amber-400" />
|
||||
<span id="add-factor-signal-title" className="text-sm font-semibold text-foreground">因子加入信号条件</span>
|
||||
</div>
|
||||
{selected && <p className="mt-1 font-mono text-[11px] text-muted">{selected.factor_name}</p>}
|
||||
</div>
|
||||
{selected && <span className="shrink-0 rounded-btn bg-elevated px-1.5 py-0.5 text-[10px] text-secondary">{selected.label}</span>}
|
||||
</div>
|
||||
|
||||
<div className="min-h-0 flex-1 space-y-3 overflow-y-auto px-4 py-3">
|
||||
{!item && (
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">选择因子（内置 / 自定义 / 复合）</span>
|
||||
<select value={selectedId} onChange={event => pickFactor(event.target.value)} className={INPUT_CLS}>
|
||||
<option value="">{library.isPending ? '加载因子中…' : '请选择因子'}</option>
|
||||
{grouped.map(([group, factors]) => (
|
||||
<optgroup key={group} label={group}>
|
||||
{factors.map(f => (
|
||||
<option key={f.id} value={f.id}>{f.label}（{f.id}）</option>
|
||||
))}
|
||||
</optgroup>
|
||||
))}
|
||||
</select>
|
||||
</label>
|
||||
)}
|
||||
{item ? (
|
||||
<div className="rounded-btn border border-border bg-base/40 px-3 py-2 text-[11px] leading-relaxed text-secondary">
|
||||
检验方向：IC {item.ic_mean == null ? '—' : (item.ic_mean * 100).toFixed(2)}%
|
||||
{icPositive ? '（值大看多，条件取高值端）' : '（值小看多，条件取低值端）'}。方向与阈值只是预填建议，可按业务调整。
|
||||
</div>
|
||||
) : (
|
||||
<div className="rounded-btn border border-border bg-base/40 px-3 py-2 text-[11px] leading-relaxed text-secondary">
|
||||
阈值按因子族惯例给建议值（RSI/KDJ 区间、z 分 ±2、有界位置 0.2/0.8、乖离 ±5%），请结合检验结果调整。
|
||||
</div>
|
||||
)}
|
||||
<div className="grid grid-cols-[5rem_1fr] gap-2">
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">运算符</span>
|
||||
<select value={op} onChange={event => pickOp(event.target.value as '>' | '<')} className={INPUT_CLS}>
|
||||
<option value=">">> 超过阈值</option>
|
||||
<option value="<">< 低于阈值</option>
|
||||
</select>
|
||||
</label>
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">阈值（按因子族惯例建议）</span>
|
||||
<input
|
||||
type="number" step="any" value={threshold}
|
||||
onChange={event => setThreshold(event.target.value)}
|
||||
className={`${INPUT_CLS} font-mono`}
|
||||
/>
|
||||
</label>
|
||||
</div>
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">信号名称</span>
|
||||
<input type="text" value={name} maxLength={24} onChange={event => setName(event.target.value)} className={INPUT_CLS} />
|
||||
</label>
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">类型</span>
|
||||
<select value={kind} onChange={event => setKind(event.target.value as 'entry' | 'exit' | 'both')} className={INPUT_CLS}>
|
||||
<option value="entry">入场</option>
|
||||
<option value="exit">出场</option>
|
||||
<option value="both">出入通用</option>
|
||||
</select>
|
||||
</label>
|
||||
<p className="text-[10px] leading-relaxed text-muted">
|
||||
保存后成为 csg_ 信号列：选股、回测、盘后监控可用；因子列由历史路径自动补算（与检验同一条计算管线），盘中实时快照无滚动窗口、该信号盘中不触发。
|
||||
</p>
|
||||
</div>
|
||||
|
||||
<div className="flex items-center justify-end gap-2 border-t border-border px-4 py-3">
|
||||
<button type="button" onClick={onClose} className="rounded-btn bg-elevated px-3 py-1.5 text-xs text-secondary transition-colors hover:text-foreground">取消</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => save.mutate()}
|
||||
disabled={save.isPending || !selected || !thresholdValid || !name.trim()}
|
||||
className="inline-flex items-center gap-1.5 rounded-btn bg-amber-500/90 px-3 py-1.5 text-xs font-medium text-base transition-colors hover:bg-amber-500 disabled:cursor-not-allowed disabled:opacity-50"
|
||||
>
|
||||
<Zap className="h-3.5 w-3.5" />
|
||||
{save.isPending ? '保存中…' : '保存信号'}
|
||||
</button>
|
||||
</div>
|
||||
</Modal>
|
||||
)
|
||||
}
|
||||
@@ -0,0 +1,195 @@
|
||||
import { useEffect, useState } from 'react'
|
||||
import { useNavigate } from 'react-router-dom'
|
||||
import { useQuery } from '@tanstack/react-query'
|
||||
import { Sparkles, X } from 'lucide-react'
|
||||
import { Modal } from '@/components/Modal'
|
||||
import { toast } from '@/components/Toast'
|
||||
import { api, type MiningAvailability, type MiningBudgetProfile } from '@/lib/api'
|
||||
import { startAutoMining } from '@/lib/miningTask'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
|
||||
const INPUT_CLS = 'h-8 w-full rounded-input border border-border bg-surface px-2.5 text-xs text-foreground focus:border-accent focus:outline-none'
|
||||
|
||||
function isoDaysAgo(days: number) {
|
||||
const value = new Date()
|
||||
value.setDate(value.getDate() - days)
|
||||
return value.toISOString().slice(0, 10)
|
||||
}
|
||||
|
||||
const PROFILE_ORDER: MiningBudgetProfile[] = ['exploratory', 'balanced', 'strict']
|
||||
const PROFILE_LABELS: Record<MiningBudgetProfile, string> = {
|
||||
exploratory: '探索',
|
||||
balanced: '均衡',
|
||||
strict: '严格',
|
||||
}
|
||||
const PROFILE_HINTS: Record<MiningBudgetProfile, string> = {
|
||||
exploratory: '门槛最低 · 只能存为待定候选',
|
||||
balanced: '门槛适中 · 推荐',
|
||||
strict: '门槛最高 · 样本要求最严',
|
||||
}
|
||||
|
||||
/** 自动挖掘入口弹窗: L1 全量统计筛选 → 达标因子池 → 嵌套样本外组合挖掘。 */
|
||||
export function AutoMiningDialog({
|
||||
onClose,
|
||||
defaultAssetType = 'stock',
|
||||
}: {
|
||||
onClose: () => void
|
||||
defaultAssetType?: 'stock' | 'etf'
|
||||
}) {
|
||||
const navigate = useNavigate()
|
||||
const [assetType, setAssetType] = useState<'stock' | 'etf'>(defaultAssetType)
|
||||
const [profile, setProfile] = useState<MiningBudgetProfile>('balanced')
|
||||
const [profileTouched, setProfileTouched] = useState(false)
|
||||
const [start, setStart] = useState(isoDaysAgo(365 * 2))
|
||||
const [end, setEnd] = useState(new Date().toISOString().slice(0, 10))
|
||||
|
||||
// 三档可用性: 数据交易日 vs 各档所需; 默认档数据不够时自动落到可用的最高档
|
||||
const availabilityArgs = {
|
||||
assetType,
|
||||
start: start || undefined,
|
||||
end: end || undefined,
|
||||
} as const
|
||||
const exploratoryQuery = useQuery({
|
||||
queryKey: QK.miningAvailability(assetType, 'exploratory', start, end),
|
||||
queryFn: () => api.miningAvailability({ ...availabilityArgs, budgetProfile: 'exploratory' }),
|
||||
staleTime: 30_000,
|
||||
})
|
||||
const balancedQuery = useQuery({
|
||||
queryKey: QK.miningAvailability(assetType, 'balanced', start, end),
|
||||
queryFn: () => api.miningAvailability({ ...availabilityArgs, budgetProfile: 'balanced' }),
|
||||
staleTime: 30_000,
|
||||
})
|
||||
const strictQuery = useQuery({
|
||||
queryKey: QK.miningAvailability(assetType, 'strict', start, end),
|
||||
queryFn: () => api.miningAvailability({ ...availabilityArgs, budgetProfile: 'strict' }),
|
||||
staleTime: 30_000,
|
||||
})
|
||||
const availability: Partial<Record<MiningBudgetProfile, MiningAvailability>> = {
|
||||
exploratory: exploratoryQuery.data,
|
||||
balanced: balancedQuery.data,
|
||||
strict: strictQuery.data,
|
||||
}
|
||||
const bars = exploratoryQuery.data?.trading_bars ?? balancedQuery.data?.trading_bars ?? strictQuery.data?.trading_bars ?? null
|
||||
const currentAvailable = availability[profile]?.eligible ?? null
|
||||
|
||||
useEffect(() => {
|
||||
if (profileTouched) return
|
||||
// 默认档 (均衡) 数据不足且存在可用档 → 自动降级一次 (用户手动改过则不再干预)。
|
||||
// 三档可用性是独立请求: 依赖须覆盖全部三档, 否则 balanced 先到、可用档后到时会错过降级。
|
||||
if (availability.balanced && !availability.balanced.eligible) {
|
||||
const fallback = PROFILE_ORDER.filter(p => availability[p]?.eligible).pop()
|
||||
if (fallback && fallback !== profile) {
|
||||
setProfile(fallback)
|
||||
toast(`当前数据 ${availability.balanced.trading_bars} 个交易日不够「均衡」档 (需 ${availability.balanced.required_bars})，已自动切到「${PROFILE_LABELS[fallback]}」档`)
|
||||
}
|
||||
}
|
||||
}, [availability.exploratory, availability.balanced, availability.strict, profile, profileTouched])
|
||||
|
||||
const submitAutoMining = () => {
|
||||
// L1 全量筛选在后端要跑几分钟: 模块级任务提交后本弹窗立刻关闭,
|
||||
// 进度/结果由挖掘页的 RunStatus + SSE 呈现 (含 0 达标原因分布)。
|
||||
void startAutoMining({
|
||||
asset_type: assetType,
|
||||
budget_profile: profile,
|
||||
start: start || null,
|
||||
end: end || null,
|
||||
})
|
||||
toast('已提交自动挖掘：正在后台筛选因子，可离开本页，进度见「挖掘」页', 'success')
|
||||
onClose()
|
||||
navigate('/factors?tab=mining')
|
||||
}
|
||||
|
||||
return (
|
||||
<Modal
|
||||
onClose={onClose}
|
||||
labelledBy="auto-mining-title"
|
||||
panelClassName="flex max-h-[86vh] w-[92vw] max-w-md flex-col overflow-hidden border border-border bg-surface shadow-xl rounded-card"
|
||||
>
|
||||
<div className="flex items-start justify-between gap-3 border-b border-border px-4 py-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<Sparkles className="h-4 w-4 shrink-0 text-accent" />
|
||||
<span id="auto-mining-title" className="text-sm font-semibold text-foreground">自动挖掘达标组合</span>
|
||||
</div>
|
||||
<p className="mt-1 text-[11px] leading-relaxed text-muted">
|
||||
不用挑因子：先全量统计筛选（|t|≥2、q≤0.1、IC/IR 达标），再用达标池做相关性去重 → 组合搜索 → 嵌套样本外验证。结果里能看到哪些因子/组合达标、不达标的差在哪。
|
||||
</p>
|
||||
</div>
|
||||
<button type="button" onClick={onClose} aria-label="关闭" className="inline-flex h-8 w-8 shrink-0 items-center justify-center text-muted transition-colors hover:text-foreground">
|
||||
<X className="h-4 w-4" />
|
||||
</button>
|
||||
</div>
|
||||
|
||||
<div className="min-h-0 flex-1 overflow-y-auto px-4 py-3">
|
||||
<div className="space-y-3">
|
||||
<div>
|
||||
<div className="mb-1 text-[10px] text-muted">资产</div>
|
||||
<div className="inline-flex rounded-btn border border-border bg-surface/80 p-0.5">
|
||||
{(['stock', 'etf'] as const).map(value => (
|
||||
<button
|
||||
key={value}
|
||||
type="button"
|
||||
onClick={() => setAssetType(value)}
|
||||
className={`h-7 rounded-[5px] px-3 text-[11px] font-medium transition-colors ${assetType === value ? 'bg-accent text-white' : 'text-secondary hover:text-foreground'}`}
|
||||
>
|
||||
{value === 'stock' ? '股票' : 'ETF'}
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
</div>
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">验证档位（决定筛选门槛与样本要求）</span>
|
||||
<select
|
||||
value={profile}
|
||||
onChange={event => { setProfileTouched(true); setProfile(event.target.value as MiningBudgetProfile) }}
|
||||
className={INPUT_CLS}
|
||||
>
|
||||
{PROFILE_ORDER.map(value => {
|
||||
const info = availability[value]
|
||||
const mark = info == null ? '' : info.eligible ? ' ✓' : ` · 需 ${info.required_bars} 日`
|
||||
return (
|
||||
<option key={value} value={value}>
|
||||
{PROFILE_LABELS[value]} · {PROFILE_HINTS[value]}{mark}
|
||||
</option>
|
||||
)
|
||||
})}
|
||||
</select>
|
||||
</label>
|
||||
<div className="grid grid-cols-2 gap-2">
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">开始</span>
|
||||
<input type="date" value={start} onChange={event => setStart(event.target.value)} className={INPUT_CLS} />
|
||||
</label>
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">结束</span>
|
||||
<input type="date" value={end} onChange={event => setEnd(event.target.value)} className={INPUT_CLS} />
|
||||
</label>
|
||||
</div>
|
||||
{currentAvailable === false && availability[profile] && (
|
||||
<div className="rounded-btn border border-danger/40 bg-danger/5 px-3 py-2 text-[11px] leading-relaxed text-danger">
|
||||
当前区间 {bars} 个交易日，不够「{PROFILE_LABELS[profile]}」档（需 {availability[profile]!.required_bars} 日）。
|
||||
{availability.exploratory?.eligible
|
||||
? '可改选「探索」档，或在数据页补充更早的历史行情后重试。'
|
||||
: '请在数据页补充更多历史行情后重试。'}
|
||||
</div>
|
||||
)}
|
||||
<p className="text-[10px] leading-relaxed text-muted">
|
||||
点击后立即转入「挖掘」页：第一步全量筛选（近一年窗口，约 1-3 分钟）在后台运行，可随意切换页面；筛选完成后挖掘任务自动开跑，结果在「最近运行」查看。
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
<div className="flex items-center justify-end gap-2 border-t border-border px-4 py-3">
|
||||
<button
|
||||
type="button"
|
||||
onClick={submitAutoMining}
|
||||
disabled={!start || !end || start > end || currentAvailable === false}
|
||||
className="inline-flex items-center gap-1.5 rounded-btn bg-accent px-3 py-1.5 text-xs font-medium text-white transition-colors hover:bg-accent/90 disabled:cursor-not-allowed disabled:opacity-50"
|
||||
>
|
||||
<Sparkles className="h-3.5 w-3.5" />
|
||||
开始自动挖掘
|
||||
</button>
|
||||
</div>
|
||||
</Modal>
|
||||
)
|
||||
}
|
||||
@@ -1,6 +1,6 @@
|
||||
import { useEffect, useMemo, useRef, useState } from 'react'
|
||||
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
|
||||
import { BookmarkPlus, ChevronRight, Clock, Layers3, ListFilter, ListPlus, Play, Search } from 'lucide-react'
|
||||
import { ArrowDown, ArrowUp, BookmarkPlus, ChevronRight, ChevronsUpDown, Clock, Info, Layers3, ListFilter, ListPlus, Play, Search, Sparkles, Zap } from 'lucide-react'
|
||||
import { DatePicker } from '@/components/DatePicker'
|
||||
import { EmptyState } from '@/components/EmptyState'
|
||||
import { toast } from '@/components/Toast'
|
||||
@@ -9,6 +9,8 @@ import { api, type FactorBatchItem, type FactorColumn } from '@/lib/api'
|
||||
import { fmtPct, priceColorClass } from '@/lib/format'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
import { FactorBacktest } from './FactorBacktest'
|
||||
import { AutoMiningDialog } from './AutoMiningDialog'
|
||||
import { AddFactorSignalDialog } from './AddFactorSignalDialog'
|
||||
import { factorBatchCandidate } from './researchCandidates'
|
||||
|
||||
const formatDate = (value: Date) => value.toISOString().slice(0, 10)
|
||||
@@ -21,13 +23,93 @@ const TODAY = formatDate(new Date())
|
||||
const INPUT_CLS = 'w-full rounded-input border border-border bg-surface px-2.5 py-1.5 text-xs focus:border-accent focus:outline-none'
|
||||
|
||||
type View = 'batch' | 'single'
|
||||
type SortKey = 'ic' | 'ir' | 'return'
|
||||
type SortKey = 'ic' | 'ir' | 'return' | 't_nw' | 'win_rate' | 'drawdown'
|
||||
type Verdict = 'valid' | 'edge' | 'invalid' | 'error'
|
||||
|
||||
function valueOrBottom(value: number | null) {
|
||||
return value == null || !Number.isFinite(value) ? Number.NEGATIVE_INFINITY : Math.abs(value)
|
||||
// P0 判读为客户端经验规则, metrics_v2 (服务端显著性) 落地后切换为服务端 verdict
|
||||
const VERDICT_META: Record<Verdict, { label: string; cls: string }> = {
|
||||
valid: { label: '有效', cls: 'bg-bull/10 text-bull' },
|
||||
edge: { label: '边缘', cls: 'bg-amber-400/10 text-amber-500' },
|
||||
invalid: { label: '无效', cls: 'bg-base text-muted' },
|
||||
error: { label: '失败', cls: 'bg-danger/10 text-danger' },
|
||||
}
|
||||
|
||||
function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void }) {
|
||||
// 服务端判读优先 (metrics_v2: NW t ≥ 2 显著 且 BH q ≤ 0.1), 缺失时降级 P0 经验规则
|
||||
const verdictOf = (item: FactorBatchItem): Verdict => {
|
||||
if (item.error) return 'error'
|
||||
const ic = Math.abs(item.ic_mean ?? 0)
|
||||
const ir = Math.abs(item.ir ?? 0)
|
||||
if (item.t_newey_west != null) {
|
||||
const significant = Math.abs(item.t_newey_west) >= 2
|
||||
&& (item.q_value == null || item.q_value <= 0.1)
|
||||
const predictive = ic >= 0.02
|
||||
if (significant && predictive) return 'valid'
|
||||
if (significant || predictive) return 'edge'
|
||||
return 'invalid'
|
||||
}
|
||||
if (ic >= 0.02 && ir >= 0.3) return 'valid'
|
||||
if (ic >= 0.02 || ir >= 0.3) return 'edge'
|
||||
return 'invalid'
|
||||
}
|
||||
|
||||
const hasServerVerdict = (results: FactorBatchItem[]): boolean =>
|
||||
results.some(item => !item.error && item.t_newey_west != null)
|
||||
|
||||
// 预设场景: 纯前端选择集, 数量按 columns 数据实时计算 (分组名缺失时该项自动为空并禁用)
|
||||
type PresetDef = { id: string; label: string; hint: string; groups?: string[]; quick?: boolean }
|
||||
const PRESETS: PresetDef[] = [
|
||||
{ id: 'all', label: '全面体检', hint: '全部因子，耗时最长' },
|
||||
{ id: 'quick', label: '快速体检', hint: '除财务组外每组各取 1 个代表因子' , quick: true },
|
||||
{ id: 'trend', label: '趋势动量', hint: '动量组 + 趋势组', groups: ['动量', '趋势'] },
|
||||
{ id: 'reversal', label: '超跌反转', hint: '超买超卖组 + 价格位置组', groups: ['超买超卖', '价格位置'] },
|
||||
{ id: 'volume', label: '量价资金', hint: '量价组 + 流动性组', groups: ['量价', '流动性'] },
|
||||
{ id: 'fundamental', label: '财务价值', hint: '财务组（需财务数据）', groups: ['财务'] },
|
||||
]
|
||||
|
||||
const INTRO_STORAGE_KEY = 'factors-intro-collapsed'
|
||||
|
||||
// 表头排序: IC/IR/t 值/多空收益按 |值| 排 (正负都是信号), 胜率/回撤按原值排
|
||||
const SORT_GETTERS: Record<SortKey, (item: FactorBatchItem) => number | null | undefined> = {
|
||||
ic: item => item.ic_mean,
|
||||
ir: item => item.ir,
|
||||
return: item => item.long_short_return,
|
||||
t_nw: item => item.t_newey_west,
|
||||
win_rate: item => item.ic_win_rate,
|
||||
drawdown: item => item.long_short_max_drawdown,
|
||||
}
|
||||
const ABS_SORT_KEYS: ReadonlySet<SortKey> = new Set(['ic', 'ir', 'return', 't_nw'])
|
||||
|
||||
function SortableTh({ label, sortKeyName, sortKey, sortAsc, onSort, className, title }: {
|
||||
label: string
|
||||
sortKeyName: SortKey
|
||||
sortKey: SortKey
|
||||
sortAsc: boolean
|
||||
onSort: (key: SortKey) => void
|
||||
className?: string
|
||||
title?: string
|
||||
}) {
|
||||
const active = sortKey === sortKeyName
|
||||
const Icon = !active ? ChevronsUpDown : sortAsc ? ArrowUp : ArrowDown
|
||||
return (
|
||||
<th
|
||||
className={`${className ?? ''} font-medium`}
|
||||
title={title}
|
||||
aria-sort={active ? (sortAsc ? 'ascending' : 'descending') : 'none'}
|
||||
>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => onSort(sortKeyName)}
|
||||
className={`inline-flex items-center gap-0.5 rounded-btn transition-colors hover:text-foreground ${active ? 'text-foreground' : ''}`}
|
||||
title={title ? `${title}${active ? '' : '（点击排序）'}` : '点击排序'}
|
||||
>
|
||||
{label}
|
||||
<Icon className={`h-3 w-3 ${active ? 'text-accent' : 'text-muted/50'}`} />
|
||||
</button>
|
||||
</th>
|
||||
)
|
||||
}
|
||||
|
||||
function BatchDiscovery({ onInspect, focusFactor }: { onInspect: (factorName: string) => void; focusFactor?: string }) {
|
||||
const queryClient = useQueryClient()
|
||||
const initialized = useRef(false)
|
||||
const [selected, setSelected] = useState<string[]>([])
|
||||
@@ -39,6 +121,10 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
const [rebalance, setRebalance] = useState<'daily' | 'weekly' | 'monthly'>('daily')
|
||||
const [fees, setFees] = useState('2')
|
||||
const [sortKey, setSortKey] = useState<SortKey>('ic')
|
||||
const [sortAsc, setSortAsc] = useState(false)
|
||||
const [factorQuery, setFactorQuery] = useState('')
|
||||
const [activePreset, setActivePreset] = useState<string | null>(null)
|
||||
const [signalFrom, setSignalFrom] = useState<FactorBatchItem | null>(null)
|
||||
|
||||
const columns = useQuery({
|
||||
queryKey: QK.factorColumns,
|
||||
@@ -63,16 +149,45 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
useEffect(() => {
|
||||
if (initialized.current || !columns.data?.columns.length) return
|
||||
initialized.current = true
|
||||
// 因子库「去检验」联动: focus 参数命中则只选该因子
|
||||
if (focusFactor && columns.data.columns.some(column => column.id === focusFactor)) {
|
||||
setSelected([focusFactor])
|
||||
setActivePreset(null)
|
||||
return
|
||||
}
|
||||
setSelected(columns.data.columns.map(column => column.id))
|
||||
}, [columns.data])
|
||||
setActivePreset('all')
|
||||
}, [columns.data, focusFactor])
|
||||
|
||||
const allColumns = useMemo(() => columns.data?.columns ?? [], [columns.data])
|
||||
const columnsByGroup = useMemo(() => {
|
||||
const groups: Record<string, FactorColumn[]> = {}
|
||||
for (const column of allColumns) (groups[column.group] ??= []).push(column)
|
||||
return groups
|
||||
}, [allColumns])
|
||||
const presetIds = (preset: PresetDef): string[] => {
|
||||
if (preset.id === 'all') return allColumns.map(item => item.id)
|
||||
if (preset.quick) {
|
||||
// 每组取列表中位代表 (上游按窗口升序排列时即窗口中位数, 如动量组取 20日)
|
||||
return Object.entries(columnsByGroup)
|
||||
.filter(([group]) => group !== '财务')
|
||||
.map(([, items]) => items[Math.floor((items.length - 1) / 2)])
|
||||
.filter(item => item != null)
|
||||
.map(item => item.id)
|
||||
}
|
||||
return (preset.groups ?? []).flatMap(group => columnsByGroup[group] ?? []).map(item => item.id)
|
||||
}
|
||||
|
||||
const factorGroups = useMemo(() => {
|
||||
// 搜索只影响展示, 不改变已选集合; 匹配 id/中文label/公式描述
|
||||
const query = factorQuery.trim().toLowerCase()
|
||||
const groups: Record<string, FactorColumn[]> = {}
|
||||
for (const column of columns.data?.columns ?? []) {
|
||||
if (query && !`${column.id} ${column.label} ${column.desc}`.toLowerCase().includes(query)) continue
|
||||
;(groups[column.group] ??= []).push(column)
|
||||
}
|
||||
return groups
|
||||
}, [columns.data])
|
||||
}, [columns.data, factorQuery])
|
||||
|
||||
const run = useMutation({
|
||||
mutationFn: () => api.factorBatch({
|
||||
@@ -93,27 +208,51 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
},
|
||||
onSuccess: () => {
|
||||
queryClient.invalidateQueries({ queryKey: QK.researchCandidates })
|
||||
toast('已保存到候选方案', 'success')
|
||||
toast('已保存到候选方案（右上角「候选方案」查看）', 'success')
|
||||
},
|
||||
onError: error => toast(`保存失败 · ${String((error as Error).message || error)}`, 'error'),
|
||||
})
|
||||
|
||||
const sortedResults = useMemo(() => {
|
||||
const values = [...(run.data?.results ?? [])]
|
||||
const getter = sortKey === 'ic'
|
||||
? (item: FactorBatchItem) => valueOrBottom(item.ic_mean)
|
||||
: sortKey === 'ir'
|
||||
? (item: FactorBatchItem) => valueOrBottom(item.ir)
|
||||
: (item: FactorBatchItem) => valueOrBottom(item.long_short_return)
|
||||
return values.sort((left, right) => getter(right) - getter(left))
|
||||
}, [run.data, sortKey])
|
||||
const getter = SORT_GETTERS[sortKey]
|
||||
const useAbs = ABS_SORT_KEYS.has(sortKey)
|
||||
const value = (item: FactorBatchItem) => {
|
||||
const raw = getter(item)
|
||||
if (raw == null || !Number.isFinite(raw)) return sortAsc ? Number.POSITIVE_INFINITY : Number.NEGATIVE_INFINITY
|
||||
return useAbs ? Math.abs(raw) : raw
|
||||
}
|
||||
return [...(run.data?.results ?? [])].sort((left, right) => (sortAsc ? value(left) - value(right) : value(right) - value(left)))
|
||||
}, [run.data, sortKey, sortAsc])
|
||||
|
||||
const applySort = (key: SortKey) => {
|
||||
if (key === sortKey) {
|
||||
setSortAsc(current => !current)
|
||||
} else {
|
||||
setSortKey(key)
|
||||
setSortAsc(false)
|
||||
}
|
||||
}
|
||||
|
||||
// 结论句: 有效数按经验规则统计; 最强按 |IC| 取 (与默认排序口径一致)
|
||||
const computableResults = useMemo(() => (run.data?.results ?? []).filter(item => !item.error), [run.data])
|
||||
const validCount = useMemo(() => computableResults.filter(item => verdictOf(item) === 'valid').length, [computableResults])
|
||||
const bestFactor = useMemo(() => {
|
||||
let best: FactorBatchItem | null = null
|
||||
for (const item of computableResults) {
|
||||
if (item.ic_mean == null) continue
|
||||
if (best == null || Math.abs(item.ic_mean) > Math.abs(best.ic_mean ?? 0)) best = item
|
||||
}
|
||||
return best
|
||||
}, [computableResults])
|
||||
|
||||
const toggleFactor = (factorName: string) => {
|
||||
setActivePreset(null)
|
||||
setSelected(current => current.includes(factorName)
|
||||
? current.filter(name => name !== factorName)
|
||||
: [...current, factorName])
|
||||
}
|
||||
const toggleGroup = (items: FactorColumn[]) => {
|
||||
setActivePreset(null)
|
||||
const ids = items.map(item => item.id)
|
||||
const allSelected = ids.every(id => selected.includes(id))
|
||||
setSelected(current => allSelected
|
||||
@@ -131,8 +270,11 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
const current = symbols.split(',').map(value => value.trim()).filter(Boolean)
|
||||
setSymbols(Array.from(new Set([...current, ...entries.map(entry => entry.symbol)])).join(','))
|
||||
}
|
||||
const allColumns = columns.data?.columns ?? []
|
||||
const allSelected = allColumns.length > 0 && allColumns.every(item => selected.includes(item.id))
|
||||
const applyPreset = (preset: PresetDef) => {
|
||||
setSelected(presetIds(preset))
|
||||
setActivePreset(preset.id)
|
||||
}
|
||||
|
||||
return (
|
||||
<div className="grid h-full min-h-0 grid-cols-1 overflow-hidden rounded-card border border-border bg-surface/80 xl:grid-cols-[18rem_minmax(0,1fr)]">
|
||||
@@ -144,14 +286,57 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
</div>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setSelected(allSelected ? [] : allColumns.map(item => item.id))}
|
||||
onClick={() => { setActivePreset(null); setSelected(allSelected ? [] : allColumns.map(item => item.id)) }}
|
||||
className="rounded-btn px-2 py-1 text-[10px] text-accent transition-colors hover:bg-accent/10"
|
||||
>
|
||||
{allSelected ? '清空' : '全选'}
|
||||
</button>
|
||||
</div>
|
||||
|
||||
<div className="space-y-2">
|
||||
<div>
|
||||
<div className="mb-1.5 text-[10px] text-muted">不知道测什么？从预设开始（一键选好因子）：</div>
|
||||
<div className="flex flex-wrap gap-1">
|
||||
{PRESETS.map(preset => {
|
||||
const ids = presetIds(preset)
|
||||
const active = activePreset === preset.id
|
||||
return (
|
||||
<button
|
||||
key={preset.id}
|
||||
type="button"
|
||||
onClick={() => applyPreset(preset)}
|
||||
disabled={ids.length === 0}
|
||||
title={preset.hint}
|
||||
className={`inline-flex items-center gap-1 rounded-btn border px-2 py-1 text-[10px] font-medium transition-colors disabled:pointer-events-none disabled:opacity-40 ${active
|
||||
? 'border-accent/60 bg-accent/10 text-accent'
|
||||
: 'border-border bg-surface text-secondary hover:border-accent/40 hover:text-accent'
|
||||
}`}
|
||||
>
|
||||
{preset.label}
|
||||
<span className="font-mono opacity-70">{ids.length}</span>
|
||||
{active && <span aria-hidden>✓</span>}
|
||||
</button>
|
||||
)
|
||||
})}
|
||||
{activePreset == null && (
|
||||
<span className="inline-flex items-center rounded-btn border border-dashed border-border px-2 py-1 text-[10px] text-muted">
|
||||
自定义
|
||||
</span>
|
||||
)}
|
||||
</div>
|
||||
</div>
|
||||
|
||||
<div className="relative">
|
||||
<Search className="pointer-events-none absolute left-2.5 top-1/2 h-3 w-3 -translate-y-1/2 text-muted" />
|
||||
<input
|
||||
type="text"
|
||||
value={factorQuery}
|
||||
onChange={event => setFactorQuery(event.target.value)}
|
||||
placeholder="搜索因子 (名称/公式)"
|
||||
className={`${INPUT_CLS} pl-7`}
|
||||
/>
|
||||
</div>
|
||||
|
||||
<div className="max-h-[45vh] space-y-2 overflow-y-auto pr-0.5">
|
||||
{Object.entries(factorGroups).map(([group, items]) => {
|
||||
const groupSelected = items.filter(item => selected.includes(item.id)).length
|
||||
return (
|
||||
@@ -182,6 +367,11 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
</div>
|
||||
)
|
||||
})}
|
||||
{Object.keys(factorGroups).length === 0 && (
|
||||
<div className="py-4 text-center text-[10px] text-muted">
|
||||
{columns.isLoading ? '因子加载中…' : '无匹配因子'}
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
|
||||
<div>
|
||||
@@ -316,27 +506,37 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
<span>{run.data.n_dates} 个交易日</span>
|
||||
<span className="inline-flex items-center gap-1"><Clock className="h-3 w-3" />{run.data.elapsed_ms.toFixed(0)} ms</span>
|
||||
</div>
|
||||
{computableResults.length > 0 && (
|
||||
<div className="mt-1 text-[11px] text-secondary">
|
||||
{computableResults.length} 个因子中 <span className="font-medium text-bull">{validCount} 个有效</span>
|
||||
<span className="text-muted">
|
||||
{hasServerVerdict(computableResults)
|
||||
? '（NW 显著性检验：|t|≥2 且 BH q≤0.1，|IC|≥0.02）'
|
||||
: '（经验规则：|IC|≥0.02 且 |IR|≥0.3）'}
|
||||
</span>
|
||||
{bestFactor && bestFactor.ic_mean != null && (
|
||||
<>。最强：<span className="font-medium text-foreground">{bestFactor.label}</span>（IC {fmtPct(bestFactor.ic_mean)}）</>
|
||||
)}
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
<div className="ml-auto self-center text-[10px] text-muted" title="点击数值列表头可切换排序键与升/降序；IC/IR/t 值/多空收益按绝对值排序（正负都是信号）。">
|
||||
点击表头排序
|
||||
</div>
|
||||
<label className="ml-auto flex items-center gap-2 text-[11px] text-muted">
|
||||
排序
|
||||
<select value={sortKey} onChange={event => setSortKey(event.target.value as SortKey)} className="h-8 rounded-input border border-border bg-surface px-2 text-xs text-secondary focus:border-accent focus:outline-none">
|
||||
<option value="ic">|IC|</option>
|
||||
<option value="ir">|IR|</option>
|
||||
<option value="return">|多空收益|</option>
|
||||
</select>
|
||||
</label>
|
||||
</div>
|
||||
<div className="overflow-x-auto">
|
||||
<table className="w-full min-w-[820px] text-xs">
|
||||
<thead className="sticky top-0 bg-elevated text-left text-[11px] text-secondary">
|
||||
<tr>
|
||||
<th className="w-12 px-3 py-2.5 text-center font-medium">排名</th>
|
||||
<th className="w-12 px-3 py-2.5 text-center font-medium" title="按当前排序键排序的名次，默认按 |IC|。">排名</th>
|
||||
<th className="px-3 py-2.5 font-medium">因子</th>
|
||||
<th className="px-3 py-2.5 text-right font-medium">IC 均值</th>
|
||||
<th className="px-3 py-2.5 text-right font-medium">IR</th>
|
||||
<th className="px-3 py-2.5 text-right font-medium">IC 胜率</th>
|
||||
<th className="px-3 py-2.5 text-right font-medium">多空收益</th>
|
||||
<th className="px-3 py-2.5 text-right font-medium">最大回撤</th>
|
||||
<SortableTh label="预测力 IC" sortKeyName="ic" sortKey={sortKey} sortAsc={sortAsc} onSort={applySort} className="px-3 py-2.5 text-right" title="每天用因子给股票打分、与次日真实涨跌算相关性（Rank IC）的均值。|IC|≥0.02 且稳定即有预测力；负值同样有效（反向使用）。" />
|
||||
<SortableTh label="稳定度 IR" sortKeyName="ir" sortKey={sortKey} sortAsc={sortAsc} onSort={applySort} className="px-3 py-2.5 text-right" title="IC 均值 ÷ IC 波动。≥0.3 值得关注，≥0.5 相当稳定。" />
|
||||
<SortableTh label="t 值(NW)" sortKeyName="t_nw" sortKey={sortKey} sortAsc={sortAsc} onSort={applySort} className="px-3 py-2.5 text-right" title="Newey-West HAC 稳健 t 值（滞后 1）：|t|≥2 视为统计显著；悬停查看多重检验校正后的 q 值（≤0.1 通过）。样本不足显示 —。" />
|
||||
<SortableTh label="预测日占比" sortKeyName="win_rate" sortKey={sortKey} sortAsc={sortAsc} onSort={applySort} className="px-3 py-2.5 text-right" title="IC 与预测方向一致的天数占比。50% 是抛硬币，55%+ 不错。" />
|
||||
<SortableTh label="多空收益" sortKeyName="return" sortKey={sortKey} sortAsc={sortAsc} onSort={applySort} className="px-3 py-2.5 text-right" title="每期买因子最高组、（模拟）卖最低组的累计收益差。A 股做空受限，此列为理论口径。" />
|
||||
<SortableTh label="最大回撤" sortKeyName="drawdown" sortKey={sortKey} sortAsc={sortAsc} onSort={applySort} className="px-3 py-2.5 text-right" title="上述多空组合最痛的一段亏损幅度，衡量拿得住不住。" />
|
||||
<th className="w-16 px-3 py-2.5 text-center font-medium" title="经验规则：|IC|≥0.02 且 |IR|≥0.3 为有效；其一达标为边缘。">结论</th>
|
||||
<th className="w-24 px-3 py-2.5 text-right font-medium">操作</th>
|
||||
</tr>
|
||||
</thead>
|
||||
@@ -345,15 +545,35 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
<tr key={item.factor_name} className="border-t border-border/70 transition-colors hover:bg-elevated/40">
|
||||
<td className="px-3 py-3 text-center font-mono text-muted">{index + 1}</td>
|
||||
<td className="px-3 py-3">
|
||||
<div className="font-medium text-foreground">{item.label}</div>
|
||||
<div className="font-medium text-foreground">
|
||||
{item.label}
|
||||
{!item.error && item.ic_mean != null && (
|
||||
item.ic_mean >= 0
|
||||
? <span className="ml-1 text-bull" title="样本内方向：值大看多（IC 为正）。历史方向不代表未来。">↑</span>
|
||||
: <span className="ml-1 text-bear" title="样本内方向：值小看多（IC 为负，反向使用）。历史方向不代表未来。">↓</span>
|
||||
)}
|
||||
</div>
|
||||
<div className="mt-0.5 text-[10px] text-muted">{item.group} · {item.factor_name}</div>
|
||||
{item.error && <div className="mt-1 text-[10px] text-danger">{item.error}</div>}
|
||||
</td>
|
||||
<td className={`px-3 py-3 text-right font-mono ${priceColorClass(item.ic_mean)}`}>{item.ic_mean == null ? '—' : fmtPct(item.ic_mean)}</td>
|
||||
<td className="px-3 py-3 text-right font-mono text-foreground">{item.ir == null ? '—' : item.ir.toFixed(2)}</td>
|
||||
<td
|
||||
className={`px-3 py-3 text-right font-mono ${item.t_newey_west != null && Math.abs(item.t_newey_west) >= 2 ? 'text-foreground' : 'text-muted'}`}
|
||||
title={item.t_newey_west != null
|
||||
? `NW t=${item.t_newey_west.toFixed(2)}${item.q_value != null ? `，BH q=${item.q_value.toFixed(3)}` : ''}（${item.q_value != null && item.q_value <= 0.1 ? '通过' : '未通过'}多重检验校正）`
|
||||
: '样本不足，无法计算'}
|
||||
>
|
||||
{item.t_newey_west == null ? '—' : item.t_newey_west.toFixed(2)}
|
||||
</td>
|
||||
<td className="px-3 py-3 text-right font-mono text-secondary">{item.ic_win_rate == null ? '—' : fmtPct(item.ic_win_rate)}</td>
|
||||
<td className={`px-3 py-3 text-right font-mono ${priceColorClass(item.long_short_return)}`}>{item.long_short_return == null ? '—' : fmtPct(item.long_short_return)}</td>
|
||||
<td className="px-3 py-3 text-right font-mono text-bear">{item.long_short_max_drawdown == null ? '—' : fmtPct(item.long_short_max_drawdown)}</td>
|
||||
<td className="px-3 py-3 text-center">
|
||||
<span className={`inline-flex rounded-btn px-1.5 py-0.5 text-[10px] font-medium ${VERDICT_META[verdictOf(item)].cls}`}>
|
||||
{VERDICT_META[verdictOf(item)].label}
|
||||
</span>
|
||||
</td>
|
||||
<td className="px-3 py-3">
|
||||
<div className="flex justify-end gap-1">
|
||||
<button
|
||||
@@ -366,6 +586,16 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
>
|
||||
<BookmarkPlus className="h-3.5 w-3.5" />
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setSignalFrom(item)}
|
||||
disabled={!!item.error}
|
||||
className="flex h-7 w-7 items-center justify-center rounded-btn text-muted transition-colors hover:bg-amber-400/10 hover:text-amber-400 disabled:opacity-40"
|
||||
title="加入信号条件（方向按 IC 预填，阈值给建议值）"
|
||||
aria-label={`把 ${item.label} 加入信号条件`}
|
||||
>
|
||||
<Zap className="h-3.5 w-3.5" />
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => onInspect(item.factor_name)}
|
||||
@@ -386,20 +616,62 @@ function BatchDiscovery({ onInspect }: { onInspect: (factorName: string) => void
|
||||
</div>
|
||||
)}
|
||||
</section>
|
||||
|
||||
{signalFrom && <AddFactorSignalDialog item={signalFrom} onClose={() => setSignalFrom(null)} />}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
export function FactorDiscovery() {
|
||||
export function FactorDiscovery({ focusFactor }: { focusFactor?: string } = {}) {
|
||||
const [view, setView] = useState<View>('batch')
|
||||
const [detailFactor, setDetailFactor] = useState('momentum_20d')
|
||||
const [autoOpen, setAutoOpen] = useState(false)
|
||||
const [introCollapsed, setIntroCollapsed] = useState(() => {
|
||||
try {
|
||||
return localStorage.getItem(INTRO_STORAGE_KEY) === '1'
|
||||
} catch {
|
||||
return false
|
||||
}
|
||||
})
|
||||
const collapseIntro = () => {
|
||||
setIntroCollapsed(true)
|
||||
try {
|
||||
localStorage.setItem(INTRO_STORAGE_KEY, '1')
|
||||
} catch { /* 隐私模式等场景忽略 */ }
|
||||
}
|
||||
const inspect = (factorName: string) => {
|
||||
setDetailFactor(factorName)
|
||||
setView('single')
|
||||
}
|
||||
return (
|
||||
<div className="flex h-full min-h-0 flex-col gap-3">
|
||||
{!introCollapsed && (
|
||||
<div className="flex shrink-0 items-start gap-2 rounded-card border border-border bg-surface/60 px-3 py-2 text-[11px] leading-relaxed text-secondary">
|
||||
<Info className="mt-0.5 h-3.5 w-3.5 shrink-0 text-accent" />
|
||||
<div className="min-w-0 flex-1">
|
||||
<span className="font-medium text-foreground">三步看懂本页：</span>
|
||||
① 因子 = 给股票打分排序的特征（如 20日涨幅、换手率变化）；② 本页检验过去哪些特征真的能预测次日涨跌；③ IC = 预测准确度（绝对值越大越准，负值反向用），IR = 稳定度（越大越稳）。
|
||||
</div>
|
||||
<button
|
||||
type="button"
|
||||
onClick={collapseIntro}
|
||||
className="shrink-0 rounded-btn px-2 py-0.5 text-[10px] text-muted transition-colors hover:bg-elevated hover:text-foreground"
|
||||
>
|
||||
知道了
|
||||
</button>
|
||||
</div>
|
||||
)}
|
||||
<div className="flex shrink-0 items-center border-b border-border/70 px-1 pb-2">
|
||||
{view === 'single' && (
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setView('batch')}
|
||||
className="mr-2 inline-flex items-center gap-1 rounded-btn px-2 py-1 text-[11px] text-muted transition-colors hover:bg-elevated hover:text-foreground"
|
||||
title="返回批量结果（筛选配置与结果保留）"
|
||||
>
|
||||
← 返回批量
|
||||
</button>
|
||||
)}
|
||||
<div className="inline-flex rounded-btn border border-border bg-surface/80 p-0.5">
|
||||
{([
|
||||
['batch', '批量筛选', ListFilter],
|
||||
@@ -419,13 +691,29 @@ export function FactorDiscovery() {
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setAutoOpen(true)}
|
||||
className="ml-auto mr-1 inline-flex items-center gap-1.5 rounded-btn border border-accent/40 bg-accent/5 px-2.5 py-1.5 text-xs font-medium text-accent transition-colors hover:bg-accent/10"
|
||||
title="不知道选什么因子？让系统全量筛选达标因子并自动搜索组合（嵌套样本外验证）"
|
||||
>
|
||||
<Sparkles className="h-3.5 w-3.5" />
|
||||
自动挖掘
|
||||
</button>
|
||||
</div>
|
||||
<div className="min-h-0 flex-1">
|
||||
{view === 'batch'
|
||||
? <BatchDiscovery onInspect={inspect} />
|
||||
: <FactorBacktest key={detailFactor} initialFactorName={detailFactor} />
|
||||
}
|
||||
{/* 批量视图保持挂载: 切到单因子检验再返回时, 选择集与结果不丢 */}
|
||||
<div className={view === 'batch' ? 'h-full' : 'hidden'}>
|
||||
<BatchDiscovery onInspect={inspect} focusFactor={focusFactor} />
|
||||
</div>
|
||||
{view === 'single' && (
|
||||
<div className="h-full">
|
||||
<FactorBacktest key={detailFactor} initialFactorName={detailFactor} />
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
|
||||
{autoOpen && <AutoMiningDialog onClose={() => setAutoOpen(false)} />}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
@@ -15,12 +15,14 @@ import {
|
||||
Rocket,
|
||||
Save,
|
||||
Settings2,
|
||||
Sparkles,
|
||||
Square,
|
||||
} from 'lucide-react'
|
||||
import { EmptyState } from '@/components/EmptyState'
|
||||
import { toast } from '@/components/Toast'
|
||||
import {
|
||||
api,
|
||||
type AutoScreening,
|
||||
type FactorColumn,
|
||||
type MiningBudgetProfile,
|
||||
type MiningCandidateGate,
|
||||
@@ -194,6 +196,62 @@ function foldKindLabel(kind?: string | null) {
|
||||
return undefined
|
||||
}
|
||||
|
||||
function pctText(value: number | null | undefined, digits = 1) {
|
||||
return typeof value === 'number' && Number.isFinite(value) ? `${(value * 100).toFixed(digits)}%` : '—'
|
||||
}
|
||||
|
||||
/** 自动挖掘 L1 筛选摘要: 达标因子清单 + 失败原因分布 (来自任务请求的 auto_screening)。 */
|
||||
function AutoScreeningCard({ screening }: { screening: AutoScreening }) {
|
||||
const gate = screening.gate
|
||||
const reasons = Object.entries(screening.reason_counts).slice(0, 5)
|
||||
const maxCount = Math.max(1, ...reasons.map(([, count]) => count))
|
||||
return (
|
||||
<section className="border-b border-border bg-accent/[0.03]">
|
||||
<div className="flex flex-wrap items-center gap-x-3 gap-y-1 px-3 py-2">
|
||||
<span className="inline-flex items-center gap-1 text-xs font-semibold text-foreground">
|
||||
<Sparkles className="h-3.5 w-3.5 text-accent" />
|
||||
自动筛选 · 达标因子池
|
||||
</span>
|
||||
<span className="font-mono text-[10px] text-secondary">
|
||||
{screening.n_qualified}/{screening.n_total} 个达标
|
||||
{screening.pool_truncated ? `（取前 ${screening.pool.length} 个入池）` : ''}
|
||||
</span>
|
||||
<span className="text-[10px] text-muted">
|
||||
门槛 |IC|≥{gate.min_abs_ic.toFixed(2)} · |IR|≥{gate.min_abs_ir.toFixed(2)} · |t|≥{gate.min_abs_t.toFixed(1)} · q≤{gate.max_q.toFixed(2)}
|
||||
，窗口 {screening.screen_window.start} ~ {screening.screen_window.end}
|
||||
</span>
|
||||
</div>
|
||||
{screening.qualified.length > 0 && (
|
||||
<div className="flex flex-wrap gap-1 px-3 pb-2">
|
||||
{screening.qualified.map(item => (
|
||||
<span
|
||||
key={item.factor_name}
|
||||
title={`${item.factor_name} · IC ${pctText(item.ic)} · IR ${item.ir?.toFixed(2) ?? '—'} · t ${item.t?.toFixed(2) ?? '—'} · q ${item.q?.toFixed(3) ?? '—'}`}
|
||||
className={`inline-flex items-center gap-1 rounded-btn border px-1.5 py-0.5 text-[10px] font-medium ${item.direction > 0 ? 'border-bull/30 bg-bull/10 text-bull' : 'border-bear/30 bg-bear/10 text-bear'}`}
|
||||
>
|
||||
{item.label}
|
||||
{item.direction > 0 ? '↑' : '↓'}
|
||||
<span className="font-mono opacity-80">{pctText(item.ic)}</span>
|
||||
</span>
|
||||
))}
|
||||
</div>
|
||||
)}
|
||||
{reasons.length > 0 && (
|
||||
<div className="space-y-1 border-t border-border/60 px-3 py-2">
|
||||
<div className="text-[10px] text-muted">未达标原因分布（{screening.n_total - screening.n_qualified} 个）：</div>
|
||||
{reasons.map(([reason, count]) => (
|
||||
<div key={reason} className="flex items-center gap-2 text-[10px]">
|
||||
<span className="w-24 shrink-0 truncate text-secondary" title={reason}>{reason}</span>
|
||||
<span className="h-1.5 rounded-full bg-accent/40" style={{ width: `${Math.max(6, (count / maxCount) * 140)}px` }} />
|
||||
<span className="font-mono text-muted">{count}</span>
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
)}
|
||||
</section>
|
||||
)
|
||||
}
|
||||
|
||||
function SummaryStrip({ result }: { result: MiningResult }) {
|
||||
const items = [
|
||||
['因子', `${result.summary.selected_factor_count}/${result.summary.factor_count}`],
|
||||
@@ -648,13 +706,13 @@ export function MiningWorkbench() {
|
||||
|
||||
<div className="sticky bottom-0 flex gap-2 bg-base/95 py-2">
|
||||
<button type="button" disabled={task.isPending || !draft.factorNames.length || (draft.strategyIds.length > 0 && strategyQuery.isLoading) || !validDateRange || availabilityQuery.isPending || availabilityQuery.isFetching || availabilityQuery.isError || !availabilityQuery.data?.eligible} onClick={runMining} className="inline-flex h-8 flex-1 items-center justify-center gap-1.5 rounded-btn bg-accent px-3 text-xs font-semibold text-white transition-opacity disabled:cursor-not-allowed disabled:opacity-50"><Play className="h-3.5 w-3.5" />开始挖掘</button>
|
||||
{task.isPending && <button type="button" title="取消任务" disabled={task.cancelling} onClick={() => void cancelMining()} className="inline-flex h-8 w-9 items-center justify-center rounded-btn border border-danger/40 text-danger hover:bg-danger/10 disabled:opacity-50"><Square className="h-3.5 w-3.5" /></button>}
|
||||
{task.isPending && <button type="button" title={task.runId ? '取消任务' : '因子筛选阶段不可取消，run 创建后可取消'} disabled={task.cancelling || !task.runId} onClick={() => void cancelMining()} className="inline-flex h-8 w-9 items-center justify-center rounded-btn border border-danger/40 text-danger hover:bg-danger/10 disabled:cursor-not-allowed disabled:opacity-50"><Square className="h-3.5 w-3.5" /></button>}
|
||||
</div>
|
||||
|
||||
<section className="border-t border-border pt-3">
|
||||
<div className="mb-2 flex items-center justify-between"><span className="text-[10px] font-semibold text-secondary">最近运行</span><button type="button" title="刷新历史" onClick={() => void runsQuery.refetch()} className="text-muted hover:text-accent"><RefreshCw className={`h-3 w-3 ${runsQuery.isFetching ? 'animate-spin' : ''}`} /></button></div>
|
||||
<div className="max-h-40 space-y-1 overflow-y-auto">
|
||||
{(runsQuery.data?.items ?? []).map(run => <button key={run.run_id} type="button" onClick={() => attachRun(run)} className={`flex w-full items-center gap-2 rounded-btn px-2 py-1.5 text-left hover:bg-elevated ${task.runId === run.run_id ? 'bg-accent/10' : ''}`}><span className={`h-1.5 w-1.5 shrink-0 rounded-full ${SUCCESS.has(run.status) ? 'bg-success' : ACTIVE.has(run.status) ? 'bg-accent' : 'bg-muted'}`} /><span className="min-w-0 flex-1 truncate font-mono text-[9px] text-secondary">{run.run_id}</span><span className="shrink-0 text-[9px] text-muted">{statusLabel(run.status)}</span></button>)}
|
||||
{(runsQuery.data?.items ?? []).map(run => <button key={run.run_id} type="button" onClick={() => attachRun(run)} className={`flex w-full items-center gap-2 rounded-btn px-2 py-1.5 text-left hover:bg-elevated ${task.runId === run.run_id ? 'bg-accent/10' : ''}`}><span className={`h-1.5 w-1.5 shrink-0 rounded-full ${SUCCESS.has(run.status) ? 'bg-success' : ACTIVE.has(run.status) ? 'bg-accent' : 'bg-muted'}`} />{run.request?.auto && <span className="shrink-0 rounded-btn bg-accent/10 px-1 text-[8px] font-medium text-accent" title="自动挖掘（因子池由统计筛选生成）">自动</span>}<span className="min-w-0 flex-1 truncate font-mono text-[9px] text-secondary">{run.run_id}</span><span className="shrink-0 text-[9px] text-muted">{statusLabel(run.status)}</span></button>)}
|
||||
{runsQuery.isError && <div className="text-[9px] text-danger">运行历史加载失败</div>}
|
||||
{!runsQuery.isLoading && !runsQuery.isError && !(runsQuery.data?.items.length) && <div className="text-[9px] text-muted">暂无持久运行</div>}
|
||||
</div>
|
||||
@@ -666,6 +724,7 @@ export function MiningWorkbench() {
|
||||
|
||||
<section className="min-w-0 bg-surface xl:max-h-[calc(100vh-9rem)] xl:overflow-y-auto">
|
||||
<RunStatus run={task.run} progress={task.progress} error={task.error} reconnecting={task.reconnecting} />
|
||||
{task.run?.request?.auto_screening && <AutoScreeningCard screening={task.run.request.auto_screening} />}
|
||||
{showingPrevious && <div className="border-b border-warning/30 bg-warning/5 px-3 py-1.5 text-[10px] text-warning">历史结果 · run {result?.run_id}。当前 run {task.runId} {task.isPending ? '仍在执行' : '未成功完成'}，以下内容仅供参考，候选操作已禁用。</div>}
|
||||
|
||||
{!result ? (
|
||||
|
||||
@@ -180,13 +180,14 @@ Object.assign(FIELD_LABEL, {
|
||||
ma20_bias: 'MA20乖离率',
|
||||
})
|
||||
const BOARD_OPTIONS = ['沪主板', '深主板', '创业板', '科创板', '北交所']
|
||||
const BASIC_FILTER_FIELDS = [
|
||||
{ key: 'price_min', label: '最低价', unit: '元' },
|
||||
{ key: 'price_max', label: '最高价', unit: '元' },
|
||||
{ key: 'amount_min', label: '最低成交额', unit: '亿', scale: 1e8 },
|
||||
{ key: 'market_cap_min', label: '最低总市值', unit: '亿', scale: 1e8 },
|
||||
{ key: 'turnover_min', label: '最低换手率', unit: '%' },
|
||||
{ key: 'turnover_max', label: '最高换手率', unit: '%' },
|
||||
// 与策略编辑器「基础参数」对齐 (engine._basic_filter_expr 支持的全部数值界),
|
||||
// 每项 min~max 成对, 面板可见即可改, 避免策略里已生效的界在回测侧不可见。
|
||||
const BASIC_FILTER_RANGES = [
|
||||
{ minKey: 'price_min', maxKey: 'price_max', label: '价格', unit: '元', step: '1' },
|
||||
{ minKey: 'float_cap_min', maxKey: 'float_cap_max', label: '流通市值', unit: '亿', scale: 1e8, step: '5' },
|
||||
{ minKey: 'market_cap_min', maxKey: 'market_cap_max', label: '总市值', unit: '亿', scale: 1e8, step: '5' },
|
||||
{ minKey: 'amount_min', maxKey: 'amount_max', label: '成交额', unit: '亿', scale: 1e8, step: '0.5' },
|
||||
{ minKey: 'turnover_min', maxKey: 'turnover_max', label: '换手率', unit: '%', step: '0.5' },
|
||||
]
|
||||
type AdvancedSettingsTab = 'params' | 'filter' | 'entry' | 'exit' | 'scoring' | 'risk' | 'range'
|
||||
type StrategyGroup = 'all' | 'custom' | 'ai' | 'builtin' | 'composite'
|
||||
@@ -589,15 +590,10 @@ const METRIC_HELP = {
|
||||
description: '回测权益从历史高点到随后最低点的最大跌幅。',
|
||||
note: '越接近 0 通常代表历史资金回撤越小。',
|
||||
},
|
||||
mcDrawdownMedian: {
|
||||
title: '蒙卡回撤中位数',
|
||||
description: '对交易收益有放回重抽样，各自计算最大回撤后取中位数。',
|
||||
note: '表示交易顺序变化时较典型的最大回撤场景。',
|
||||
},
|
||||
mcDrawdown95: {
|
||||
title: '蒙卡回撤 95% 边界',
|
||||
description: '交易收益重抽样结果中偏悲观的最大回撤边界。',
|
||||
note: '约有 95% 的模拟顺序回撤不劣于此值，但不是未来承诺。',
|
||||
mcDrawdown: {
|
||||
title: '蒙卡回撤 (中位/95%)',
|
||||
description: '对交易收益有放回重抽样 1000 次, 各算最大回撤: 中位值 = 典型运气下的回撤, 95% 边界 = 偏悲观的压力边界。',
|
||||
note: '中位值用于和实际最大回撤对照判断序列运气; 资金管理按 95% 边界准备。交易数多时两者可能同时逼近 -100%, 参考价值下降。',
|
||||
},
|
||||
tradeCount: {
|
||||
title: '交易数',
|
||||
@@ -663,11 +659,13 @@ function MetricLabel({ label, metric }: { label: string; metric: MetricHelpKey }
|
||||
}
|
||||
|
||||
function Stat({ label, value, color }: { label: ReactNode; value: string; color?: string }) {
|
||||
// 长值 (如蒙卡回撤双值) 降一档字号, 保证单行不撑高卡片
|
||||
const compact = value.length > 12
|
||||
return (
|
||||
<div className="min-w-0 rounded-btn border border-border/70 bg-elevated/70 px-3 py-2">
|
||||
<div className="text-[11px] text-secondary">{label}</div>
|
||||
<div
|
||||
className="mt-1 break-words text-sm font-mono font-semibold leading-tight tracking-tight num xl:text-base"
|
||||
className={`mt-1 break-words font-mono font-semibold leading-tight tracking-tight num ${compact ? 'text-xs xl:text-sm' : 'text-sm xl:text-base'}`}
|
||||
style={{ color: color ?? 'inherit' }}
|
||||
title={value}
|
||||
>
|
||||
@@ -1008,7 +1006,16 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
onLoadConsumed?.()
|
||||
// eslint-disable-next-line react-hooks/exhaustive-deps -- 仅在切换候选时执行一次性回填
|
||||
}, [loadCandidate])
|
||||
const [resultTab, setResultTab] = useState<'daily' | 'trades' | 'picks'>('daily')
|
||||
const [resultTab, setResultTab] = useState<'daily' | 'trades' | 'picks' | 'attribution'>('daily')
|
||||
// 因子归因表的 id → 中文标签 (字段视图 + 因子库合并, 自定义因子也在库内)
|
||||
const factorMetaQ = useQuery({ queryKey: QK.factorColumns, queryFn: api.factorColumns, staleTime: 300_000 })
|
||||
const factorLibQ = useQuery({ queryKey: QK.factorLibrary('all'), queryFn: () => api.factorLibrary(), staleTime: 60_000 })
|
||||
const factorLabels = useMemo(() => {
|
||||
const m = new Map<string, string>()
|
||||
for (const c of factorMetaQ.data?.columns ?? []) m.set(c.id, c.label)
|
||||
for (const f of factorLibQ.data?.factors ?? []) if (!m.has(f.id)) m.set(f.id, f.label)
|
||||
return m
|
||||
}, [factorMetaQ.data, factorLibQ.data])
|
||||
const [dailyPage, setDailyPage] = useState(0)
|
||||
const [tradePage, setTradePage] = useState(0)
|
||||
const [tradePageSize, setTradePageSize] = useState(10)
|
||||
@@ -2255,11 +2262,17 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
<Stat label={<MetricLabel label="索提诺" metric="sortino" />} value={pick('sortino') != null ? Number(pick('sortino')).toFixed(2) : '—'} />
|
||||
<Stat label={<MetricLabel label="最大回撤" metric="maxDrawdown" />} value={pick('max_drawdown') != null ? fmtPct(pick('max_drawdown') as number) : '—'}
|
||||
color="#34d399" />
|
||||
<Stat label={<MetricLabel label="蒙卡回撤(中位)" metric="mcDrawdownMedian" />} value={pick('mc_maxdd_p50') != null ? fmtPct(pick('mc_maxdd_p50') as number) : '—'}
|
||||
color="#34d399" />
|
||||
<Stat label={<MetricLabel label="蒙卡回撤(95%边界)" metric="mcDrawdown95" />} value={pick('mc_maxdd_p95') != null ? fmtPct(pick('mc_maxdd_p95') as number) : '—'}
|
||||
color="#34d399" />
|
||||
<Stat
|
||||
label={<MetricLabel label="蒙卡回撤 中位/95%" metric="mcDrawdown" />}
|
||||
value={`${pick('mc_maxdd_p50') != null ? fmtPct(pick('mc_maxdd_p50') as number) : '—'}/${pick('mc_maxdd_p95') != null ? fmtPct(pick('mc_maxdd_p95') as number) : '—'}`}
|
||||
color="#34d399"
|
||||
/>
|
||||
<Stat label={<MetricLabel label="胜率" metric="winRate" />} value={pick('win_rate') != null ? fmtPct(pick('win_rate') as number) : '—'} />
|
||||
<Stat
|
||||
label={<MetricLabel label="盈亏比" metric="profitFactor" />}
|
||||
value={pick('profit_factor') != null ? Number(pick('profit_factor')).toFixed(2) : '—'}
|
||||
color={pick('profit_factor') != null ? statValueColor(Number(pick('profit_factor')) - 1) : undefined}
|
||||
/>
|
||||
<Stat label={<MetricLabel label="交易数" metric="tradeCount" />} value={pick('n_trades') != null ? String(pick('n_trades')) : '—'} />
|
||||
{result.stats.full_kind === 'candidate_execution' ? (
|
||||
<Stat label={<MetricLabel label="平均持仓" metric="avgDuration" />} value={pick('avg_duration') != null ? `${Number(pick('avg_duration')).toFixed(1)}天` : '—'} />
|
||||
@@ -2312,11 +2325,14 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
</div>
|
||||
)}
|
||||
|
||||
{/* Tab: 按日期 / 交易明细 / 选股分析 */}
|
||||
{(result.trades.length > 0 || result.per_symbol_stats.length > 0) && (
|
||||
{/* Tab: 按日期 / 交易明细 / 选股分析 / 因子归因 */}
|
||||
{(result.trades.length > 0 || result.per_symbol_stats.length > 0 || (result.factor_attribution?.factors.length ?? 0) > 0) && (
|
||||
<div className="rounded-card border border-border overflow-hidden">
|
||||
<div className="flex items-center gap-1 border-b border-border px-4 pt-2">
|
||||
{(['daily', 'trades', 'picks'] as const).map(t => (
|
||||
{(['daily', 'trades', 'picks', 'attribution'] as const).map(t => {
|
||||
const attributionCount = result.factor_attribution?.factors.length ?? 0
|
||||
if (t === 'attribution' && attributionCount === 0) return null
|
||||
return (
|
||||
<button
|
||||
key={t}
|
||||
onClick={() => setResultTab(t)}
|
||||
@@ -2330,9 +2346,12 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
? `每日交易 (${dailyTradeRows.length})`
|
||||
: t === 'trades'
|
||||
? `交易明细 (${sortedTrades.length})`
|
||||
: `选股分析 (${result.per_symbol_stats.length})`}
|
||||
: t === 'picks'
|
||||
? `选股分析 (${result.per_symbol_stats.length})`
|
||||
: `因子归因 (${attributionCount})`}
|
||||
</button>
|
||||
))}
|
||||
)
|
||||
})}
|
||||
</div>
|
||||
|
||||
{resultTab === 'daily' && (
|
||||
@@ -2576,6 +2595,55 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
</tbody>
|
||||
</table>
|
||||
)}
|
||||
|
||||
{/* 因子归因: 入场信号日因子值 × 成交盈亏 */}
|
||||
{resultTab === 'attribution' && result.factor_attribution && (
|
||||
<div className="px-4 py-3">
|
||||
<div className="mb-1 flex items-center justify-between">
|
||||
<span className="text-xs font-medium text-secondary">入场信号日因子均值</span>
|
||||
<span className="text-[10px] text-muted">
|
||||
胜单 {result.factor_attribution.n_win} · 败单 {result.factor_attribution.n_lose}
|
||||
</span>
|
||||
</div>
|
||||
<p className="mb-2 text-[10px] leading-4 text-muted">
|
||||
对比盈利单与亏损单入场时的因子取值：胜单均值明显高于败单 → 该因子在本轮交易里贡献了正筛选力；反之在拖后腿。原始因子量纲不同，只看相对差异。
|
||||
</p>
|
||||
<div className="overflow-x-auto">
|
||||
<table className="w-full text-xs">
|
||||
<thead>
|
||||
<tr className="border-b border-border text-[10px] text-muted">
|
||||
<th className="px-2 py-1.5 text-left font-normal">因子</th>
|
||||
<th className="px-2 py-1.5 text-right font-normal">胜单均值</th>
|
||||
<th className="px-2 py-1.5 text-right font-normal">败单均值</th>
|
||||
<th className="px-2 py-1.5 text-right font-normal">差值(胜-败)</th>
|
||||
<th className="px-2 py-1.5 text-right font-normal">样本(胜/败)</th>
|
||||
</tr>
|
||||
</thead>
|
||||
<tbody>
|
||||
{result.factor_attribution.factors.map(f => {
|
||||
const diff = f.win_mean != null && f.lose_mean != null ? f.win_mean - f.lose_mean : null
|
||||
return (
|
||||
<tr key={f.factor} className="border-b border-border/50">
|
||||
<td className="px-2 py-1.5 font-mono text-foreground">
|
||||
{f.factor}
|
||||
{factorLabels.get(f.factor) && (
|
||||
<span className="ml-1 font-sans text-muted">{factorLabels.get(f.factor)}</span>
|
||||
)}
|
||||
</td>
|
||||
<td className="px-2 py-1.5 text-right font-mono text-bull">{f.win_mean == null ? '—' : f.win_mean}</td>
|
||||
<td className="px-2 py-1.5 text-right font-mono text-bear">{f.lose_mean == null ? '—' : f.lose_mean}</td>
|
||||
<td className={`px-2 py-1.5 text-right font-mono ${diff == null ? 'text-muted' : diff >= 0 ? 'text-bull' : 'text-bear'}`}>
|
||||
{diff == null ? '—' : (diff >= 0 ? '+' : '') + diff.toFixed(4)}
|
||||
</td>
|
||||
<td className="px-2 py-1.5 text-right font-mono text-secondary">{f.win_n}/{f.lose_n}</td>
|
||||
</tr>
|
||||
)
|
||||
})}
|
||||
</tbody>
|
||||
</table>
|
||||
</div>
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
)}
|
||||
|
||||
@@ -2703,19 +2771,36 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
启用基础过滤
|
||||
</label>
|
||||
<div className="grid grid-cols-1 gap-3 sm:grid-cols-2">
|
||||
{BASIC_FILTER_FIELDS.map(field => {
|
||||
{BASIC_FILTER_RANGES.map(field => {
|
||||
const scale = field.scale ?? 1
|
||||
const raw = basicFilter[field.key]
|
||||
const bound = (key: string) => {
|
||||
const raw = basicFilter[key]
|
||||
return raw == null ? null : Number(raw) / scale
|
||||
}
|
||||
const setBound = (key: string) => (n: number | null) =>
|
||||
updateBasicFilter(key, n == null ? null : n * scale)
|
||||
return (
|
||||
<label key={field.key} className="block">
|
||||
<label key={field.minKey} className="block">
|
||||
<span className="mb-1 block text-[11px] text-secondary">{field.label}({field.unit})</span>
|
||||
<div className="flex items-center gap-1.5">
|
||||
<NumberField
|
||||
value={raw == null ? null : Number(raw) / scale}
|
||||
value={bound(field.minKey)}
|
||||
min={0}
|
||||
step={field.unit === '%' ? 0.1 : 0.01}
|
||||
onChange={n => updateBasicFilter(field.key, n == null ? null : n * scale)}
|
||||
step={Number(field.step)}
|
||||
placeholder="不限"
|
||||
onChange={setBound(field.minKey)}
|
||||
className={INPUT_CLS}
|
||||
/>
|
||||
<span className="text-[11px] text-muted">~</span>
|
||||
<NumberField
|
||||
value={bound(field.maxKey)}
|
||||
min={0}
|
||||
step={Number(field.step)}
|
||||
placeholder="不限"
|
||||
onChange={setBound(field.maxKey)}
|
||||
className={INPUT_CLS}
|
||||
/>
|
||||
</div>
|
||||
</label>
|
||||
)
|
||||
})}
|
||||
@@ -2728,6 +2813,9 @@ export function StrategyBacktest({ loadCandidate, onLoadConsumed }: {
|
||||
/>
|
||||
排除 ST / 退市
|
||||
</label>
|
||||
<div className="text-[11px] leading-5 text-muted">
|
||||
字段与策略编辑器「基础参数」一致，初始值取自策略文件；清空某项即改为不限（会覆盖策略原值）。
|
||||
</div>
|
||||
<div className="flex flex-wrap gap-1.5">
|
||||
{BOARD_OPTIONS.map(board => {
|
||||
const boards = Array.isArray(basicFilter.boards) ? basicFilter.boards : []
|
||||
|
||||
@@ -0,0 +1,192 @@
|
||||
import { useMemo, useState } from 'react'
|
||||
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
|
||||
import { Combine, Save, Search } from 'lucide-react'
|
||||
import { toast } from '@/components/Toast'
|
||||
import { api } from '@/lib/api'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
|
||||
const INPUT_CLS = 'w-full rounded-input border border-border bg-surface px-2.5 py-1.5 text-xs focus:border-accent focus:outline-none'
|
||||
const MAX_MEMBERS = 8
|
||||
|
||||
type WeightMode = 'manual' | 'equal'
|
||||
|
||||
export function FactorComposite() {
|
||||
const [query, setQuery] = useState('')
|
||||
const [members, setMembers] = useState<Record<string, number>>({})
|
||||
const [mode, setMode] = useState<WeightMode>('equal')
|
||||
const [label, setLabel] = useState('')
|
||||
const [saveId, setSaveId] = useState('')
|
||||
|
||||
const queryClient = useQueryClient()
|
||||
const lib = useQuery({
|
||||
queryKey: QK.factorLibrary('all'),
|
||||
queryFn: () => api.factorLibrary(),
|
||||
})
|
||||
const save = useMutation({
|
||||
mutationFn: () => api.factorCompositeCreate({
|
||||
id: saveId.trim() || undefined,
|
||||
label: label.trim(),
|
||||
members,
|
||||
}),
|
||||
onSuccess: data => {
|
||||
toast(`已注册复合因子 ${data.id} (v${data.version})，策略评分中可直接引用`, 'success')
|
||||
void queryClient.invalidateQueries({ queryKey: ['factors-library'] })
|
||||
setMembers({})
|
||||
setLabel('')
|
||||
setSaveId('')
|
||||
},
|
||||
onError: (error: Error) => toast(`保存失败 · ${error.message}`, 'error'),
|
||||
})
|
||||
|
||||
const candidates = useMemo(() => {
|
||||
const keyword = query.trim().toLowerCase()
|
||||
return (lib.data?.factors ?? []).filter(item =>
|
||||
item.kind === 'base' || item.kind === 'virtual'
|
||||
).filter(item =>
|
||||
!keyword || `${item.id} ${item.label}`.toLowerCase().includes(keyword)
|
||||
)
|
||||
}, [lib.data, query])
|
||||
|
||||
const memberList = Object.entries(members)
|
||||
const effectiveWeights = mode === 'equal'
|
||||
? Object.fromEntries(memberList.map(([id]) => [id, 1 / Math.max(memberList.length, 1)]))
|
||||
: members
|
||||
|
||||
const toggleMember = (id: string) => {
|
||||
setMembers(current => {
|
||||
const next = { ...current }
|
||||
if (id in next) delete next[id]
|
||||
else if (Object.keys(next).length < MAX_MEMBERS) next[id] = 1
|
||||
return next
|
||||
})
|
||||
}
|
||||
const setWeight = (id: string, weight: number) => {
|
||||
setMembers(current => ({ ...current, [id]: weight }))
|
||||
}
|
||||
const weightSum = Object.values(effectiveWeights).reduce((left, right) => left + right, 0)
|
||||
|
||||
const labelOf = (id: string) =>
|
||||
(lib.data?.factors ?? []).find(item => item.id === id)?.label ?? id
|
||||
|
||||
return (
|
||||
<div className="grid h-full min-h-0 grid-cols-1 gap-3 overflow-hidden xl:grid-cols-[20rem_minmax(0,1fr)]">
|
||||
<section className="flex min-h-0 flex-col gap-2 overflow-hidden rounded-card border border-border bg-surface/80 p-3">
|
||||
<div className="flex items-center gap-2">
|
||||
<Search className="h-3.5 w-3.5 text-muted" />
|
||||
<input
|
||||
type="text"
|
||||
value={query}
|
||||
onChange={event => setQuery(event.target.value)}
|
||||
placeholder="搜索因子加入组合"
|
||||
className={INPUT_CLS}
|
||||
/>
|
||||
</div>
|
||||
<div className="min-h-0 flex-1 space-y-1 overflow-y-auto pr-0.5">
|
||||
{lib.isLoading && <div className="py-4 text-center text-[10px] text-muted">因子库加载中…</div>}
|
||||
{candidates.map(item => {
|
||||
const selected = item.id in members
|
||||
return (
|
||||
<button
|
||||
key={item.id}
|
||||
type="button"
|
||||
onClick={() => toggleMember(item.id)}
|
||||
disabled={!selected && memberList.length >= MAX_MEMBERS}
|
||||
className={`flex w-full items-center justify-between gap-2 rounded-btn border px-2 py-1.5 text-left text-[11px] transition-colors disabled:opacity-40 ${selected
|
||||
? 'border-accent/60 bg-accent/10 text-accent'
|
||||
: 'border-border bg-surface text-secondary hover:border-accent/40'
|
||||
}`}
|
||||
>
|
||||
<span className="min-w-0 flex-1 truncate">
|
||||
{item.label}
|
||||
<span className="ml-1 font-mono text-[9px] text-muted">{item.id}</span>
|
||||
</span>
|
||||
{selected && <span aria-hidden>✓</span>}
|
||||
</button>
|
||||
)
|
||||
})}
|
||||
</div>
|
||||
<div className="border-t border-border pt-1.5 text-[10px] text-muted">
|
||||
已选 {memberList.length}/{MAX_MEMBERS} 个成员；点击已选成员可移除。
|
||||
</div>
|
||||
</section>
|
||||
|
||||
<section className="flex min-h-0 flex-col gap-3 overflow-y-auto rounded-card border border-border bg-surface/80 p-3">
|
||||
<div className="flex items-center gap-2">
|
||||
<Combine className="h-4 w-4 text-accent" />
|
||||
<span className="text-sm font-medium text-foreground">组合因子构建器</span>
|
||||
<span className="text-[10px] text-muted">值 = Σ 权重 × 截面 z 分(成员)；注册后策略评分可选</span>
|
||||
</div>
|
||||
|
||||
{memberList.length === 0 ? (
|
||||
<div className="rounded-btn border border-dashed border-border px-3 py-8 text-center text-xs text-muted">
|
||||
从左侧选择 2~{MAX_MEMBERS} 个因子开始构建组合。
|
||||
</div>
|
||||
) : (
|
||||
<>
|
||||
<div className="flex items-center gap-1.5 text-[11px]">
|
||||
<span className="text-muted">权重方式</span>
|
||||
{(['equal', 'manual'] as WeightMode[]).map(value => (
|
||||
<button
|
||||
key={value}
|
||||
type="button"
|
||||
onClick={() => setMode(value)}
|
||||
className={`rounded-btn border px-2 py-0.5 font-medium transition-colors ${mode === value
|
||||
? 'border-accent/60 bg-accent/10 text-accent'
|
||||
: 'border-border text-secondary hover:text-accent'
|
||||
}`}
|
||||
>
|
||||
{value === 'equal' ? '均等' : '手动'}
|
||||
</button>
|
||||
))}
|
||||
{mode === 'manual' && (
|
||||
<span className="text-muted">当前权重和 {weightSum.toFixed(2)}（按比例归一使用）</span>
|
||||
)}
|
||||
</div>
|
||||
<div className="space-y-1.5">
|
||||
{memberList.map(([id, rawWeight]) => (
|
||||
<div key={id} className="flex items-center gap-2 rounded-btn border border-border bg-base/40 px-2.5 py-1.5">
|
||||
<span className="min-w-0 flex-1 truncate text-xs text-foreground">
|
||||
{labelOf(id)} <span className="font-mono text-[10px] text-muted">{id}</span>
|
||||
</span>
|
||||
{mode === 'manual' ? (
|
||||
<>
|
||||
<input
|
||||
type="number"
|
||||
step="0.1"
|
||||
min="0.1"
|
||||
value={rawWeight}
|
||||
onChange={event => setWeight(id, Math.max(0.1, Number(event.target.value) || 0.1))}
|
||||
className="w-20 rounded-input border border-border bg-surface px-2 py-1 text-right font-mono text-xs"
|
||||
aria-label={`${labelOf(id)} 权重`}
|
||||
/>
|
||||
<button type="button" onClick={() => toggleMember(id)} className="text-[10px] text-muted hover:text-danger">移除</button>
|
||||
</>
|
||||
) : (
|
||||
<span className="font-mono text-xs text-secondary">{(1 / memberList.length).toFixed(3)}</span>
|
||||
)}
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
<div className="flex flex-wrap items-center gap-1.5">
|
||||
<input type="text" value={label} onChange={event => setLabel(event.target.value)} placeholder="组合名称 (必填)" className={`${INPUT_CLS} h-8 w-40`} />
|
||||
<input type="text" value={saveId} onChange={event => setSaveId(event.target.value)} placeholder="id (可空, cf_ 前缀)" className={`${INPUT_CLS} h-8 w-44 font-mono`} />
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => save.mutate()}
|
||||
disabled={!label.trim() || memberList.length < 2 || save.isPending}
|
||||
className="inline-flex h-8 items-center gap-1.5 rounded-btn bg-accent px-3 text-xs font-medium text-white transition-colors hover:bg-accent/90 disabled:cursor-not-allowed disabled:opacity-50"
|
||||
>
|
||||
<Save className="h-3.5 w-3.5" />
|
||||
{save.isPending ? '保存中…' : '注册复合因子'}
|
||||
</button>
|
||||
</div>
|
||||
<div className="text-[10px] leading-relaxed text-muted">
|
||||
注册后复合因子进入注册表（因子库可见）；在策略编辑的评分配置中直接引用即可完成研究线→交易线接入。
|
||||
「检验」页可像普通因子一样检验组合。ICIR 自动权重模式将随 metrics 历史数据积累后开放。
|
||||
</div>
|
||||
</>
|
||||
)}
|
||||
</section>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
@@ -0,0 +1,558 @@
|
||||
import { useEffect, useMemo, useRef, useState } from 'react'
|
||||
import { useNavigate } from 'react-router-dom'
|
||||
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
|
||||
import { CircleAlert, CircleCheck, FlaskConical, PenLine, Play, Save, ShieldQuestion } from 'lucide-react'
|
||||
import { toast } from '@/components/Toast'
|
||||
import { api, type FactorTrialResponse, type FactorValidateResponse } from '@/lib/api'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
|
||||
const INPUT_CLS = 'w-full rounded-input border border-border bg-surface px-2.5 py-1.5 text-xs focus:border-accent focus:outline-none'
|
||||
const CHIP_CLS = 'shrink-0 rounded-btn border border-border bg-base/60 px-1.5 py-0.5 font-mono text-[10px] text-secondary transition-colors hover:border-accent/40 hover:text-accent cursor-pointer'
|
||||
|
||||
const DEFAULT_FORMULA = 'rank(-ts_sum(change_pct, 5))'
|
||||
|
||||
const TEMPLATES: { label: string; formula: string; note: string }[] = [
|
||||
{ label: '5日反转', formula: 'rank(-ts_sum(change_pct, 5))', note: '近 5 日累计涨幅的截面倒数' },
|
||||
{ label: '量价相关', formula: 'ts_corr(close / ts_delay(close, 1) - 1, volume, 20)', note: '日收益与成交量的 20 日滚动相关' },
|
||||
{ label: '波动变化', formula: 'ts_std(close / ts_delay(close, 1) - 1, 20)', note: '20 日滚动日收益波动' },
|
||||
{ label: '换手异动', formula: 'ts_zscore(turnover_rate, 60)', note: '换手率相对自身 60 日分布的 z 分' },
|
||||
{ label: '乖离组合', formula: 'zscore(close / ma20 - 1) - zscore(ts_mean(turnover_rate, 5))', note: '价格乖离与换手均值的截面差' },
|
||||
]
|
||||
|
||||
/** 后端 dsl.OPERATORS 的全部 25 个算子, 按家族分组; 点击插入, 悬停显示签名与约束。 */
|
||||
const OPERATOR_GROUPS: { label: string; ops: { name: string; desc: string }[] }[] = [
|
||||
{
|
||||
label: '时序', ops: [
|
||||
{ name: 'ts_mean', desc: 'ts_mean(x,n) 滚动均值, n∈[2,512]' },
|
||||
{ name: 'ts_sum', desc: 'ts_sum(x,n) 滚动求和, n∈[2,512]' },
|
||||
{ name: 'ts_std', desc: 'ts_std(x,n) 滚动标准差, n∈[2,512]' },
|
||||
{ name: 'ts_max', desc: 'ts_max(x,n) 滚动最大值, n∈[2,512]' },
|
||||
{ name: 'ts_min', desc: 'ts_min(x,n) 滚动最小值, n∈[2,512]' },
|
||||
{ name: 'ts_delay', desc: 'ts_delay(x,n) n 期前的值, n≤512 (禁止负数)' },
|
||||
{ name: 'ts_delta', desc: 'ts_delta(x,n) n 期差分 (x − n 期前的值)' },
|
||||
{ name: 'ts_rank', desc: 'ts_rank(x,n) 滚动分位' },
|
||||
{ name: 'ts_zscore', desc: 'ts_zscore(x,n) 滚动 z 分' },
|
||||
{ name: 'ts_corr', desc: 'ts_corr(x,y,n) 滚动相关系数' },
|
||||
{ name: 'ts_cov', desc: 'ts_cov(x,y,n) 滚动协方差' },
|
||||
{ name: 'ts_quantile', desc: 'ts_quantile(x,n,q) 滚动分位, q∈(0,1)' },
|
||||
{ name: 'decay_linear', desc: 'decay_linear(x,n) 线性衰减加权均值 (近端权重大)' },
|
||||
],
|
||||
},
|
||||
{
|
||||
label: '截面', ops: [
|
||||
{ name: 'rank', desc: 'rank(x) 当日横截面百分位' },
|
||||
{ name: 'zscore', desc: 'zscore(x) 当日横截面 z 分' },
|
||||
{ name: 'winsorize', desc: 'winsorize(x,k) 截面截尾至 μ±kσ, k∈[1,6] 默认 3' },
|
||||
],
|
||||
},
|
||||
{
|
||||
label: '工具', ops: [
|
||||
{ name: 'if_else', desc: 'if_else(c,a,b) 条件选择' },
|
||||
{ name: 'min', desc: 'min(a,b) 两值取小' },
|
||||
{ name: 'max', desc: 'max(a,b) 两值取大' },
|
||||
{ name: 'log', desc: 'log(x) 自然对数' },
|
||||
{ name: 'abs', desc: 'abs(x) 绝对值' },
|
||||
{ name: 'sign', desc: 'sign(x) 符号函数' },
|
||||
{ name: 'sqrt', desc: 'sqrt(x) 平方根' },
|
||||
{ name: 'power', desc: 'power(x,c) 幂运算, |c|≤4' },
|
||||
{ name: 'clamp', desc: 'clamp(x,lo,hi) 截断到 [lo,hi]' },
|
||||
],
|
||||
},
|
||||
]
|
||||
const OPERATOR_COUNT = OPERATOR_GROUPS.reduce((n, g) => n + g.ops.length, 0)
|
||||
|
||||
interface EditingFactor {
|
||||
id: string
|
||||
label: string
|
||||
group: string
|
||||
formula: string
|
||||
description: string
|
||||
direction: string
|
||||
version: number
|
||||
}
|
||||
|
||||
/** 因子编辑器: 写公式 → 校验 → 试算 → 保存; 支持编辑已有自定义因子 (版本提升)。 */
|
||||
export function FactorEditor({ editId = '' }: { editId?: string }) {
|
||||
const navigate = useNavigate()
|
||||
const queryClient = useQueryClient()
|
||||
const [formula, setFormula] = useState(DEFAULT_FORMULA)
|
||||
const [assetType, setAssetType] = useState<'stock' | 'etf'>('stock')
|
||||
const [validation, setValidation] = useState<FactorValidateResponse | null>(null)
|
||||
const [trial, setTrial] = useState<FactorTrialResponse | null>(null)
|
||||
const [saveLabel, setSaveLabel] = useState('')
|
||||
const [saveId, setSaveId] = useState('')
|
||||
const [saveGroup, setSaveGroup] = useState('自定义')
|
||||
const [saveDescription, setSaveDescription] = useState('')
|
||||
const [direction, setDirection] = useState<'none' | 'high' | 'low'>('none')
|
||||
const [editing, setEditing] = useState<EditingFactor | null>(null)
|
||||
const [savedId, setSavedId] = useState('')
|
||||
const [savedVersion, setSavedVersion] = useState<number | null>(null)
|
||||
const textareaRef = useRef<HTMLTextAreaElement>(null)
|
||||
|
||||
const library = useQuery({
|
||||
queryKey: QK.factorLibrary('all'),
|
||||
queryFn: () => api.factorLibrary(),
|
||||
staleTime: 30_000,
|
||||
})
|
||||
const columns = useQuery({
|
||||
queryKey: QK.factorColumns,
|
||||
queryFn: api.factorColumns,
|
||||
staleTime: 300_000,
|
||||
})
|
||||
const customFactors = (library.data?.factors ?? []).filter(item => item.kind === 'custom')
|
||||
const fieldGroups = useMemo(() => {
|
||||
const cols = columns.data?.columns ?? []
|
||||
const groups: Record<string, typeof cols> = {}
|
||||
for (const col of cols) (groups[col.group] ??= []).push(col)
|
||||
return Object.entries(groups).sort(([a], [b]) => a.localeCompare(b, 'zh'))
|
||||
}, [columns.data])
|
||||
|
||||
const loadForEdit = (id: string) => {
|
||||
const item = customFactors.find(f => f.id === id)
|
||||
if (!item) return
|
||||
setEditing({
|
||||
id: item.id, label: item.label, group: item.group,
|
||||
formula: item.formula, description: '', direction: item.direction,
|
||||
version: item.version,
|
||||
})
|
||||
setFormula(item.formula)
|
||||
setSaveLabel(item.label)
|
||||
setSaveGroup(item.group)
|
||||
setSaveId(item.id)
|
||||
setDirection((item.direction === 'high' || item.direction === 'low') ? item.direction : 'none')
|
||||
setValidation(null)
|
||||
setTrial(null)
|
||||
setSavedId('')
|
||||
setSavedVersion(null)
|
||||
}
|
||||
|
||||
// URL ?edit= 透传 (因子库「编辑公式」入口)
|
||||
useEffect(() => {
|
||||
if (editId && customFactors.length && !editing) {
|
||||
if (customFactors.some(f => f.id === editId)) loadForEdit(editId)
|
||||
}
|
||||
// eslint-disable-next-line react-hooks/exhaustive-deps
|
||||
}, [editId, customFactors.length])
|
||||
|
||||
const validate = useMutation({
|
||||
mutationFn: () => api.factorValidate(formula),
|
||||
onSuccess: data => setValidation(data),
|
||||
})
|
||||
const runTrial = useMutation({
|
||||
mutationFn: () => api.factorTrial({ formula, asset_type: assetType, days: 40 }),
|
||||
onSuccess: data => setTrial(data),
|
||||
})
|
||||
const save = useMutation({
|
||||
mutationFn: () => {
|
||||
const payload = {
|
||||
label: saveLabel.trim(),
|
||||
group: saveGroup.trim() || '自定义',
|
||||
formula,
|
||||
description: saveDescription.trim(),
|
||||
direction,
|
||||
}
|
||||
return editing
|
||||
? api.factorCustomUpdate(editing.id, payload)
|
||||
: api.factorCustomCreate({ ...payload, id: saveId.trim() || undefined })
|
||||
},
|
||||
onSuccess: data => {
|
||||
setSavedId(data.id)
|
||||
setSavedVersion(data.version)
|
||||
if (editing) setEditing({ ...editing, version: data.version })
|
||||
toast(editing
|
||||
? `${data.id} 已更新到 v${data.version}${'status' in data && data.status === 'draft' ? ' (公式变化, 状态回草稿)' : ''}`
|
||||
: `已注册自定义因子 ${data.id} (v${data.version})，可在因子库与检验页使用`, 'success')
|
||||
void queryClient.invalidateQueries({ queryKey: ['factors-library'] })
|
||||
},
|
||||
onError: (error: Error) => toast(`保存失败 · ${error.message}`, 'error'),
|
||||
})
|
||||
|
||||
const runBoth = () => {
|
||||
setTrial(null)
|
||||
validate.mutate()
|
||||
}
|
||||
const canTrial = validation?.ok === true && !runTrial.isPending
|
||||
const formulaDirty = editing != null && editing.formula !== formula
|
||||
|
||||
// 点击错误 → 聚焦并选中出错位置附近的片段
|
||||
const locateError = (offset: number | undefined) => {
|
||||
if (offset == null || !textareaRef.current) return
|
||||
const start = Math.max(0, Math.min(offset, formula.length - 1))
|
||||
const end = Math.min(formula.length, start + 12)
|
||||
textareaRef.current.focus()
|
||||
textareaRef.current.setSelectionRange(start, end)
|
||||
}
|
||||
|
||||
// 算子/字段 chip → 插入到公式光标处 (无光标则追加到末尾)
|
||||
const insertSnippet = (snippet: string) => {
|
||||
const el = textareaRef.current
|
||||
if (!el) {
|
||||
setFormula(formula + snippet)
|
||||
return
|
||||
}
|
||||
const start = el.selectionStart ?? formula.length
|
||||
const end = el.selectionEnd ?? start
|
||||
const next = formula.slice(0, start) + snippet + formula.slice(end)
|
||||
setFormula(next)
|
||||
setValidation(null)
|
||||
setTrial(null)
|
||||
setSavedId('')
|
||||
requestAnimationFrame(() => {
|
||||
el.focus()
|
||||
const pos = start + snippet.length
|
||||
el.setSelectionRange(pos, pos)
|
||||
})
|
||||
}
|
||||
|
||||
// 模板/我的因子整体替换公式; 非初始内容时先确认, 防误触丢失半成品
|
||||
const applyTemplate = (next: string) => {
|
||||
const untouched = !editing && (formula === DEFAULT_FORMULA || formula.trim() === '')
|
||||
if (!untouched && !window.confirm('当前公式将被整体替换, 继续?')) return
|
||||
setFormula(next)
|
||||
setValidation(null)
|
||||
setTrial(null)
|
||||
setSavedId('')
|
||||
}
|
||||
|
||||
return (
|
||||
<div className="grid h-full min-h-0 grid-cols-1 gap-3 overflow-hidden xl:grid-cols-2">
|
||||
{/* ── 左栏: 编辑 ── */}
|
||||
<section className="flex min-h-0 flex-col gap-3 rounded-card border border-border bg-surface/80 p-3 xl:overflow-y-auto">
|
||||
<div className="flex flex-wrap items-center gap-2">
|
||||
<FlaskConical className="h-4 w-4 shrink-0 text-accent" />
|
||||
<span className="shrink-0 text-sm font-medium text-foreground">因子编辑器</span>
|
||||
{editing && (
|
||||
<span className="inline-flex shrink-0 items-center gap-1 rounded-btn border border-accent/40 bg-accent/5 px-1.5 py-0.5 text-[10px] font-medium text-accent">
|
||||
<PenLine className="h-3 w-3" />
|
||||
编辑 {editing.id} · 当前 v{editing.version}
|
||||
</span>
|
||||
)}
|
||||
{customFactors.length > 0 && (
|
||||
<select
|
||||
value={editing?.id ?? ''}
|
||||
onChange={event => {
|
||||
if (!event.target.value) { setEditing(null); return }
|
||||
loadForEdit(event.target.value)
|
||||
}}
|
||||
className="ml-auto h-7 max-w-[14rem] shrink-0 rounded-input border border-border bg-surface px-1.5 text-[10px] text-secondary"
|
||||
aria-label="编辑已有自定义因子"
|
||||
>
|
||||
<option value="">编辑已有因子…</option>
|
||||
{customFactors.map(item => (
|
||||
<option key={item.id} value={item.id}>{item.label} ({item.id} v{item.version})</option>
|
||||
))}
|
||||
</select>
|
||||
)}
|
||||
</div>
|
||||
|
||||
<div>
|
||||
<div className="mb-1.5 flex flex-wrap items-center justify-between gap-2">
|
||||
<label className="shrink-0 text-xs font-medium text-secondary">公式 (DSL)</label>
|
||||
<select
|
||||
className="h-7 shrink-0 rounded-input border border-border bg-surface px-1.5 text-[10px] text-secondary"
|
||||
onChange={event => {
|
||||
const value = event.target.value
|
||||
if (!value) return
|
||||
const mine = value.startsWith('my:') ? customFactors.find(f => f.id === value.slice(3)) : undefined
|
||||
const template = TEMPLATES.find(item => item.label === value)
|
||||
const next = mine?.formula ?? template?.formula
|
||||
if (next != null) applyTemplate(next)
|
||||
event.target.value = ''
|
||||
}}
|
||||
aria-label="从模板开始"
|
||||
>
|
||||
<option value="">从模板开始 / 我的因子…</option>
|
||||
<optgroup label="经典模板">
|
||||
{TEMPLATES.map(template => (
|
||||
<option key={template.label} value={template.label}>{template.label} · {template.note}</option>
|
||||
))}
|
||||
</optgroup>
|
||||
{customFactors.length > 0 && (
|
||||
<optgroup label={`我的因子 (${customFactors.length})`}>
|
||||
{customFactors.map(item => (
|
||||
<option key={item.id} value={`my:${item.id}`}>{item.label} ({item.id})</option>
|
||||
))}
|
||||
</optgroup>
|
||||
)}
|
||||
</select>
|
||||
</div>
|
||||
<textarea
|
||||
ref={textareaRef}
|
||||
value={formula}
|
||||
onChange={event => { setFormula(event.target.value); setValidation(null); setTrial(null); setSavedId('') }}
|
||||
spellCheck={false}
|
||||
rows={5}
|
||||
className="w-full resize-y rounded-input border border-border bg-base/60 px-3 py-2 font-mono text-xs leading-relaxed text-foreground focus:border-accent focus:outline-none"
|
||||
placeholder="例如: rank(-ts_sum(change_pct, 5))"
|
||||
/>
|
||||
</div>
|
||||
|
||||
{/* 算子/字段点选插入: 写公式不用背名字 */}
|
||||
<div className="space-y-1.5">
|
||||
<div>
|
||||
<div className="mb-1 text-[10px] text-muted">算子 ({OPERATOR_COUNT}) · 点击插入 · 悬停看签名与约束</div>
|
||||
<div className="flex flex-wrap items-center gap-x-3 gap-y-1" role="group" aria-label="插入算子">
|
||||
{OPERATOR_GROUPS.map(group => (
|
||||
<div key={group.label} className="flex flex-wrap items-center gap-1">
|
||||
<span className="shrink-0 text-[10px] text-muted">{group.label}</span>
|
||||
{group.ops.map(op => (
|
||||
<button key={op.name} type="button" title={op.desc} onClick={() => insertSnippet(`${op.name}(`)} className={CHIP_CLS}>
|
||||
{op.name}
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
</div>
|
||||
<details className="rounded-btn border border-border bg-base/40 px-2.5 py-1.5 text-[10px] text-muted">
|
||||
<summary className="cursor-pointer text-secondary">可用字段 ({columns.data?.columns.length ?? 0}) · 点击插入</summary>
|
||||
<div className="mt-1.5 max-h-40 space-y-1.5 overflow-y-auto">
|
||||
{fieldGroups.map(([group, cols]) => (
|
||||
<div key={group}>
|
||||
<div className="mb-0.5 text-[10px] text-muted">{group}</div>
|
||||
<div className="flex flex-wrap gap-1">
|
||||
{cols.map(col => (
|
||||
<button
|
||||
key={col.id}
|
||||
type="button"
|
||||
title={`${col.label} — ${col.desc}`}
|
||||
onClick={() => insertSnippet(col.id)}
|
||||
className={CHIP_CLS}
|
||||
>
|
||||
<span className="font-mono">{col.id}</span>
|
||||
<span className="ml-1 font-sans text-muted">{col.label}</span>
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
</div>
|
||||
))}
|
||||
{columns.isPending && <div className="text-muted">加载字段中…</div>}
|
||||
</div>
|
||||
</details>
|
||||
</div>
|
||||
|
||||
{/* 动作行: nowrap 防止按钮文字折行 */}
|
||||
<div className="flex flex-wrap items-center gap-2">
|
||||
<select value={assetType} onChange={event => setAssetType(event.target.value as typeof assetType)} className={`${INPUT_CLS} h-8 w-24 shrink-0`} aria-label="试算资产">
|
||||
<option value="stock">股票</option>
|
||||
<option value="etf">ETF</option>
|
||||
</select>
|
||||
<button
|
||||
type="button"
|
||||
onClick={runBoth}
|
||||
disabled={!formula.trim() || validate.isPending}
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 whitespace-nowrap rounded-btn border border-border bg-surface px-3 text-xs font-medium text-secondary transition-colors hover:border-accent/40 hover:text-accent disabled:cursor-not-allowed disabled:opacity-50"
|
||||
>
|
||||
<ShieldQuestion className="h-3.5 w-3.5" />
|
||||
{validate.isPending ? '校验中…' : '校验'}
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => runTrial.mutate()}
|
||||
disabled={!canTrial}
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 whitespace-nowrap rounded-btn bg-accent px-3 text-xs font-medium text-white transition-colors hover:bg-accent/90 disabled:cursor-not-allowed disabled:opacity-50"
|
||||
title={validation?.ok ? '用最近 40 个交易日数据试算 IC' : '请先校验通过'}
|
||||
>
|
||||
<Play className="h-3.5 w-3.5" />
|
||||
{runTrial.isPending ? '试算中…' : '试算 40 日'}
|
||||
</button>
|
||||
</div>
|
||||
|
||||
{validate.isError && (
|
||||
<div className="rounded-btn border border-danger/30 bg-danger/10 px-3 py-2 text-xs text-danger">
|
||||
{String((validate.error as Error).message)}
|
||||
</div>
|
||||
)}
|
||||
</section>
|
||||
|
||||
{/* ── 右栏: 校验 / 试算 / 保存 ── */}
|
||||
<section className="flex min-h-0 flex-col gap-3 rounded-card border border-border bg-surface/80 p-3 xl:overflow-y-auto">
|
||||
<div className="text-sm font-medium text-foreground">校验与试算</div>
|
||||
|
||||
{!validation && !validate.isPending && (
|
||||
<div className="rounded-btn border border-dashed border-border px-3 py-6 text-center text-xs text-muted">
|
||||
<div className="mb-2">写好公式后按顺序走完四步：</div>
|
||||
<div className="flex flex-wrap items-center justify-center gap-x-1.5 gap-y-1">
|
||||
<span className="rounded-btn bg-elevated px-2 py-1">1 写公式</span>
|
||||
<span>→</span>
|
||||
<span className="rounded-btn bg-elevated px-2 py-1">2 校验语法</span>
|
||||
<span>→</span>
|
||||
<span className="rounded-btn bg-elevated px-2 py-1">3 试算 IC</span>
|
||||
<span>→</span>
|
||||
<span className="rounded-btn bg-elevated px-2 py-1">4 保存草稿</span>
|
||||
</div>
|
||||
<div className="mt-2 text-[10px]">算子/字段在左侧点击插入; 不知从何写起可从模板或「我的因子」开始; 校验错误可点击定位到字符。</div>
|
||||
</div>
|
||||
)}
|
||||
|
||||
{validation && (
|
||||
<div className="space-y-2">
|
||||
{validation.ok ? (
|
||||
<div className="flex items-center gap-2 rounded-btn border border-bull/30 bg-bull/10 px-3 py-2 text-xs text-bull">
|
||||
<CircleCheck className="h-4 w-4 shrink-0" />
|
||||
公式有效。预热 {validation.warmup_bars} 个交易日
|
||||
{validation.cross_sectional && '，含截面算子。'}
|
||||
</div>
|
||||
) : (
|
||||
<div className="space-y-1.5 rounded-btn border border-danger/30 bg-danger/10 px-3 py-2">
|
||||
<div className="flex items-center gap-2 text-xs font-medium text-danger">
|
||||
<CircleAlert className="h-4 w-4 shrink-0" />公式未通过校验 (点击错误可定位):
|
||||
</div>
|
||||
{validation.errors.map((error, index) => (
|
||||
<button
|
||||
key={index}
|
||||
type="button"
|
||||
onClick={() => locateError(error.position?.offset)}
|
||||
className="block w-full text-left font-mono text-[10px] leading-relaxed text-danger transition-colors hover:text-danger/80"
|
||||
title={error.position != null ? `定位到第 ${error.position.offset} 个字符附近` : undefined}
|
||||
>
|
||||
[{error.code}] {error.message}
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
)}
|
||||
{validation.ok && (
|
||||
<div className="grid grid-cols-1 gap-2 text-[10px] sm:grid-cols-2">
|
||||
<div className="rounded-btn border border-border bg-base/40 px-2.5 py-1.5">
|
||||
<div className="mb-0.5 text-muted">依赖列 (递归展开)</div>
|
||||
<code className="break-all font-mono text-secondary">{validation.dependencies.join(', ') || '—'}</code>
|
||||
</div>
|
||||
<div className="rounded-btn border border-border bg-base/40 px-2.5 py-1.5">
|
||||
<div className="mb-0.5 text-muted">引用因子</div>
|
||||
<code className="break-all font-mono text-secondary">{validation.referenced_factors.join(', ') || '—'}</code>
|
||||
</div>
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
)}
|
||||
|
||||
{runTrial.isPending && (
|
||||
<div className="flex items-center gap-3 rounded-btn border border-accent/30 bg-accent/5 px-3 py-2.5 text-xs text-secondary">
|
||||
<span className="h-4 w-4 animate-spin rounded-full border-2 border-accent/25 border-t-accent" />
|
||||
正在加载面板并试算最近 40 个交易日…
|
||||
</div>
|
||||
)}
|
||||
{runTrial.isError && (
|
||||
<div className="rounded-btn border border-danger/30 bg-danger/10 px-3 py-2 text-xs text-danger">
|
||||
{String((runTrial.error as Error).message)}
|
||||
</div>
|
||||
)}
|
||||
|
||||
{trial && !runTrial.isPending && (
|
||||
<div className="space-y-2">
|
||||
{trial.n_dates === 0 ? (
|
||||
<div className="rounded-btn border border-border bg-base/40 px-3 py-2 text-xs text-muted">
|
||||
{trial.message ?? '无有效 IC 截面。'}
|
||||
</div>
|
||||
) : (
|
||||
<>
|
||||
<div className="grid grid-cols-3 gap-2 text-center sm:grid-cols-5">
|
||||
{([
|
||||
['IC 均值', trial.ic_mean?.toFixed(4) ?? '—'],
|
||||
['ICIR', trial.ir?.toFixed(2) ?? '—'],
|
||||
['IC 胜率', trial.ic_win_rate != null ? `${(trial.ic_win_rate * 100).toFixed(0)}%` : '—'],
|
||||
['t值(NW)', trial.t_newey_west != null ? trial.t_newey_west.toFixed(2) : '—'],
|
||||
['空值率', trial.null_ratio != null ? `${(trial.null_ratio * 100).toFixed(0)}%` : '—'],
|
||||
] as [string, string][]).map(([label, value]) => (
|
||||
<div key={label} className="rounded-btn border border-border bg-base/40 px-1 py-1.5" title={label === 't值(NW)' ? 'Newey-West 稳健 t 值 (滞后1), |t|≥2 视为显著; 40 日样本仅作参考' : undefined}>
|
||||
<div className="whitespace-nowrap text-[10px] text-muted">{label}</div>
|
||||
<div className="font-mono text-sm text-foreground">{value}</div>
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
<div className="rounded-btn border border-border bg-base/40 px-3 py-2">
|
||||
<div className="mb-1.5 text-[10px] text-muted">IC 走势 (最近 {trial.ic_series.length} 日)</div>
|
||||
<div className="flex h-16 items-center gap-px">
|
||||
{trial.ic_series.map(point => {
|
||||
const height = Math.min(Math.abs(point.ic) * 160, 100)
|
||||
return (
|
||||
<div key={point.date} className="flex h-full flex-1 flex-col justify-center" title={`${point.date}: ${point.ic} (${point.n_symbols}只)`}>
|
||||
<div className="flex h-1/2 items-end"><div className="w-full rounded-t-sm bg-bull/70" style={{ height: point.ic > 0 ? `${height / 2}%` : 0 }} /></div>
|
||||
<div className="flex h-1/2 items-start"><div className="w-full rounded-b-sm bg-bear/70" style={{ height: point.ic < 0 ? `${height / 2}%` : 0 }} /></div>
|
||||
</div>
|
||||
)
|
||||
})}
|
||||
</div>
|
||||
</div>
|
||||
<div className="text-[10px] leading-relaxed text-muted">
|
||||
试算仅为快照预览 (无成本/分层), 完整检验请到「检验」页运行。样本内表现不代表未来。
|
||||
</div>
|
||||
|
||||
{/* 注册/更新区: 试算有非空输出才可保存 (服务端同样 fail-closed 校验) */}
|
||||
{(editing || (trial.ok && trial.n_dates > 0)) && (
|
||||
<div className="rounded-btn border border-border bg-base/40 p-2.5">
|
||||
<div className="mb-1.5 text-[11px] font-medium text-secondary">
|
||||
{editing ? `更新 ${editing.id} (保存为新版本 v${editing.version + 1})` : '注册为自定义因子'}
|
||||
</div>
|
||||
<div className="grid grid-cols-1 gap-1.5 sm:grid-cols-2">
|
||||
<input
|
||||
type="text"
|
||||
value={saveLabel}
|
||||
onChange={event => { setSaveLabel(event.target.value); setSavedId('') }}
|
||||
placeholder="名称 (必填, 如 我的反转)"
|
||||
className={`${INPUT_CLS} h-8`}
|
||||
/>
|
||||
<input
|
||||
type="text"
|
||||
value={editing ? saveGroup : saveId}
|
||||
onChange={event => {
|
||||
if (editing) setSaveGroup(event.target.value)
|
||||
else setSaveId(event.target.value)
|
||||
}}
|
||||
placeholder={editing ? '分组' : 'id (可空, 自动生成 uf_ 前缀)'}
|
||||
className={`${INPUT_CLS} h-8 font-mono`}
|
||||
/>
|
||||
</div>
|
||||
<div className="mt-1.5 flex flex-wrap items-center gap-1.5">
|
||||
<select value={direction} onChange={event => setDirection(event.target.value as typeof direction)} className={`${INPUT_CLS} h-8 w-32 shrink-0`} aria-label="预期方向">
|
||||
<option value="none">方向未知</option>
|
||||
<option value="high">值大看多</option>
|
||||
<option value="low">值小看多</option>
|
||||
</select>
|
||||
<input
|
||||
type="text"
|
||||
value={saveDescription}
|
||||
onChange={event => setSaveDescription(event.target.value)}
|
||||
placeholder="描述 (可空, 记录设计意图, 如: 短线超跌反转)"
|
||||
className={`${INPUT_CLS} h-8 min-w-[12rem] flex-1`}
|
||||
maxLength={500}
|
||||
/>
|
||||
</div>
|
||||
<div className="mt-1.5 flex flex-wrap items-center gap-2">
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => save.mutate()}
|
||||
disabled={!saveLabel.trim() || save.isPending}
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 whitespace-nowrap rounded-btn bg-accent px-3 text-xs font-medium text-white transition-colors hover:bg-accent/90 disabled:cursor-not-allowed disabled:opacity-50"
|
||||
>
|
||||
<Save className="h-3.5 w-3.5" />
|
||||
{save.isPending ? '保存中…' : editing ? (formulaDirty ? '保存新版本 (回草稿)' : '保存元数据') : '保存 (草稿态)'}
|
||||
</button>
|
||||
{savedId && !save.isPending && (
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => navigate(`/factors?tab=inspect&focus=${savedId}`)}
|
||||
className="inline-flex h-8 shrink-0 items-center gap-1.5 whitespace-nowrap rounded-btn border border-accent/40 px-3 text-xs font-medium text-accent transition-colors hover:bg-accent/10"
|
||||
>
|
||||
去检验 {savedId}
|
||||
</button>
|
||||
)}
|
||||
</div>
|
||||
<div className="mt-1 text-[10px] text-muted">
|
||||
{editing
|
||||
? '公式变化会保存为新版本并回草稿态 (需重新检验后激活); 仅改名称/分组保留当前状态。'
|
||||
: '保存后状态为 draft；到「检验」页跑完整检验确认有效后，在因子库中激活。'}
|
||||
</div>
|
||||
{savedId && savedVersion != null && (
|
||||
<div className="mt-1 text-[10px] text-bull">已保存: {savedId} v{savedVersion}</div>
|
||||
)}
|
||||
</div>
|
||||
)}
|
||||
</>
|
||||
)}
|
||||
</div>
|
||||
)}
|
||||
</section>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
@@ -0,0 +1,383 @@
|
||||
import { useMemo, useState } from 'react'
|
||||
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
|
||||
import { ArrowRight, FlaskConical, PenLine, Search, Sparkles, Trash2, X } from 'lucide-react'
|
||||
import { Modal } from '@/components/Modal'
|
||||
import { toast } from '@/components/Toast'
|
||||
import { api, type FactorLibraryItem } from '@/lib/api'
|
||||
import { QK } from '@/lib/queryKeys'
|
||||
import { GenerateFactorStrategyDialog } from './GenerateFactorStrategyDialog'
|
||||
|
||||
const INPUT_CLS = 'rounded-input border border-border bg-surface px-2.5 py-1.5 text-xs focus:border-accent focus:outline-none'
|
||||
|
||||
const KIND_META: Record<FactorLibraryItem['kind'], { label: string; cls: string }> = {
|
||||
base: { label: '基础', cls: 'bg-elevated text-secondary' },
|
||||
virtual: { label: '虚拟', cls: 'bg-accent/10 text-accent' },
|
||||
composite: { label: '复合', cls: 'bg-bull/10 text-bull' },
|
||||
custom: { label: '自定义', cls: 'bg-amber-400/10 text-amber-500' },
|
||||
}
|
||||
|
||||
export function FactorLibrary({ onInspect, onEdit }: { onInspect: (factorId: string) => void; onEdit?: (factorId: string) => void }) {
|
||||
const [query, setQuery] = useState('')
|
||||
const [kind, setKind] = useState('all')
|
||||
const [group, setGroup] = useState('all')
|
||||
const [detail, setDetail] = useState<FactorLibraryItem | null>(null)
|
||||
|
||||
const lib = useQuery({
|
||||
queryKey: QK.factorLibrary('all'),
|
||||
queryFn: () => api.factorLibrary(),
|
||||
})
|
||||
const factors = lib.data?.factors ?? []
|
||||
|
||||
const groups = useMemo(
|
||||
() => Array.from(new Set(factors.map(item => item.group))),
|
||||
[factors],
|
||||
)
|
||||
|
||||
const filtered = useMemo(() => {
|
||||
const keyword = query.trim().toLowerCase()
|
||||
return factors.filter(item => {
|
||||
if (kind !== 'all' && item.kind !== kind) return false
|
||||
if (group !== 'all' && item.group !== group) return false
|
||||
if (keyword && !`${item.id} ${item.label} ${item.formula}`.toLowerCase().includes(keyword)) return false
|
||||
return true
|
||||
})
|
||||
}, [factors, query, kind, group])
|
||||
|
||||
return (
|
||||
<div className="flex h-full min-h-0 flex-col overflow-hidden rounded-card border border-border bg-surface/80">
|
||||
<div className="flex flex-wrap items-center gap-2 border-b border-border px-3 py-2.5">
|
||||
<div className="relative">
|
||||
<Search className="pointer-events-none absolute left-2.5 top-1/2 h-3 w-3 -translate-y-1/2 text-muted" />
|
||||
<input
|
||||
type="text"
|
||||
value={query}
|
||||
onChange={event => setQuery(event.target.value)}
|
||||
placeholder="搜索因子 (id / 名称 / 公式)"
|
||||
className={`${INPUT_CLS} w-56 pl-7`}
|
||||
/>
|
||||
</div>
|
||||
<select value={kind} onChange={event => setKind(event.target.value)} className={`${INPUT_CLS} w-24`} aria-label="因子类型">
|
||||
<option value="all">全部类型</option>
|
||||
<option value="base">基础</option>
|
||||
<option value="virtual">虚拟</option>
|
||||
<option value="composite">复合</option>
|
||||
<option value="custom">自定义</option>
|
||||
</select>
|
||||
<select value={group} onChange={event => setGroup(event.target.value)} className={`${INPUT_CLS} w-28`} aria-label="因子分组">
|
||||
<option value="all">全部分组</option>
|
||||
{groups.map(name => <option key={name} value={name}>{name}</option>)}
|
||||
</select>
|
||||
<span className="ml-auto text-[10px] text-muted">{lib.isLoading ? '加载中…' : `${filtered.length} / ${factors.length} 个因子`}</span>
|
||||
</div>
|
||||
|
||||
<div className="min-h-0 flex-1 overflow-auto">
|
||||
{lib.isError && (
|
||||
<div className="m-3 rounded-btn border border-danger/30 bg-danger/10 px-3 py-2 text-sm text-danger">
|
||||
{String((lib.error as Error).message)}
|
||||
</div>
|
||||
)}
|
||||
{lib.isLoading && (
|
||||
<div className="p-6 text-center text-xs text-muted">因子库加载中…</div>
|
||||
)}
|
||||
{!lib.isLoading && !lib.isError && filtered.length === 0 && (
|
||||
<div className="p-6 text-center text-xs text-muted">无匹配因子</div>
|
||||
)}
|
||||
{!lib.isLoading && filtered.length > 0 && (
|
||||
<table className="w-full min-w-[760px] text-xs">
|
||||
<thead className="sticky top-0 bg-elevated text-left text-[11px] text-secondary">
|
||||
<tr>
|
||||
<th className="px-3 py-2.5 font-medium">因子</th>
|
||||
<th className="px-3 py-2.5 font-medium">分组</th>
|
||||
<th className="px-3 py-2.5 font-medium">类型</th>
|
||||
<th className="px-3 py-2.5 font-medium">公式</th>
|
||||
<th className="px-3 py-2.5 font-medium" title="按需计算所需的最少历史交易日数">预热</th>
|
||||
<th className="px-3 py-2.5 font-medium">适用</th>
|
||||
<th className="w-24 px-3 py-2.5 text-right font-medium">操作</th>
|
||||
</tr>
|
||||
</thead>
|
||||
<tbody>
|
||||
{filtered.map(item => (
|
||||
<tr
|
||||
key={item.id}
|
||||
onClick={() => setDetail(item)}
|
||||
className="cursor-pointer border-t border-border/70 transition-colors hover:bg-elevated/40"
|
||||
>
|
||||
<td className="px-3 py-2.5">
|
||||
<div className="font-medium text-foreground">{item.label}</div>
|
||||
<div className="mt-0.5 font-mono text-[10px] text-muted">{item.id}</div>
|
||||
</td>
|
||||
<td className="px-3 py-2.5 text-secondary">{item.group}</td>
|
||||
<td className="px-3 py-2.5">
|
||||
<span className={`inline-flex rounded-btn px-1.5 py-0.5 text-[10px] font-medium ${KIND_META[item.kind].cls}`}>
|
||||
{KIND_META[item.kind].label}
|
||||
</span>
|
||||
{item.pit && <span className="ml-1 text-[10px] text-amber-500" title="点时数据: 仅使用公告日不晚于当日的财务数据">点时</span>}
|
||||
</td>
|
||||
<td className="max-w-[22rem] px-3 py-2.5">
|
||||
<span className="block truncate text-secondary" title={item.formula}>{item.formula}</span>
|
||||
</td>
|
||||
<td className="px-3 py-2.5 font-mono text-secondary">{item.warmup_bars > 1 ? `${item.warmup_bars}日` : '—'}</td>
|
||||
<td className="px-3 py-2.5 text-muted">
|
||||
{item.asset_types.includes('stock') && item.asset_types.includes('etf') ? '股票/ETF' : item.asset_types.join('/')}
|
||||
</td>
|
||||
<td className="px-3 py-2.5 text-right">
|
||||
<button
|
||||
type="button"
|
||||
onClick={event => { event.stopPropagation(); onInspect(item.id) }}
|
||||
className="inline-flex items-center gap-1 rounded-btn px-2 py-1 text-[10px] text-accent transition-colors hover:bg-accent/10"
|
||||
title="去检验该因子（切到检验页并只选它）"
|
||||
>
|
||||
去检验 <ArrowRight className="h-3 w-3" />
|
||||
</button>
|
||||
</td>
|
||||
</tr>
|
||||
))}
|
||||
</tbody>
|
||||
</table>
|
||||
)}
|
||||
</div>
|
||||
|
||||
{detail && <FactorDetailModal item={detail} onClose={() => setDetail(null)} onInspect={onInspect} onEdit={onEdit} />}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
function FactorDetailModal({
|
||||
item,
|
||||
onClose,
|
||||
onInspect,
|
||||
onEdit,
|
||||
}: {
|
||||
item: FactorLibraryItem
|
||||
onClose: () => void
|
||||
onInspect: (factorId: string) => void
|
||||
onEdit?: (factorId: string) => void
|
||||
}) {
|
||||
const queryClient = useQueryClient()
|
||||
const setStatus = useMutation({
|
||||
mutationFn: (status: 'draft' | 'active' | 'watch' | 'retired') => api.factorSetStatus(item.id, status),
|
||||
onSuccess: data => {
|
||||
toast(`${item.label} 状态已更新为 ${data.status}`, 'success')
|
||||
void queryClient.invalidateQueries({ queryKey: ['factors-library'] })
|
||||
onClose()
|
||||
},
|
||||
onError: (error: Error) => toast(`状态更新失败 · ${error.message}`, 'error'),
|
||||
})
|
||||
const isDynamic = item.kind === 'custom' || item.kind === 'composite'
|
||||
const [groupDraft, setGroupDraft] = useState(item.group)
|
||||
const [confirmDelete, setConfirmDelete] = useState(false)
|
||||
const [blockedRefs, setBlockedRefs] = useState<string[] | null>(null)
|
||||
const [generateOpen, setGenerateOpen] = useState(false)
|
||||
const removeFactor = useMutation({
|
||||
mutationFn: (force: boolean) => api.factorDelete(item.id, force),
|
||||
onSuccess: data => {
|
||||
const refs = data.removed_references ?? []
|
||||
toast(
|
||||
refs.length > 0
|
||||
? `${item.label} 已强制删除 · ${refs.length} 处引用需手动调整 (${refs.join('、')})`
|
||||
: `${item.label} 已删除`,
|
||||
'success',
|
||||
)
|
||||
void queryClient.invalidateQueries({ queryKey: ['factors-library'] })
|
||||
onClose()
|
||||
},
|
||||
onError: (error: Error) => {
|
||||
// 409 时 detail 为 {message, references} → request() 已 stringify 成 JSON 文本
|
||||
try {
|
||||
const parsed = JSON.parse(error.message) as { references?: string[] }
|
||||
if (Array.isArray(parsed.references) && parsed.references.length > 0) {
|
||||
setBlockedRefs(parsed.references)
|
||||
return
|
||||
}
|
||||
} catch { /* 非 JSON 报错, 走通用提示 */ }
|
||||
toast(`删除失败 · ${error.message}`, 'error')
|
||||
},
|
||||
})
|
||||
const setGroup = useMutation({
|
||||
mutationFn: () => api.factorSetGroup(item.id, groupDraft.trim()),
|
||||
onSuccess: data => {
|
||||
toast(`${item.label} 分组已改为 ${data.group}`, 'success')
|
||||
void queryClient.invalidateQueries({ queryKey: ['factors-library'] })
|
||||
onClose()
|
||||
},
|
||||
onError: (error: Error) => toast(`分组修改失败 · ${error.message}`, 'error'),
|
||||
})
|
||||
const rows: [string, string][] = [
|
||||
['标识', item.id],
|
||||
['名称', item.label],
|
||||
['分组', item.group],
|
||||
['类型', KIND_META[item.kind].label],
|
||||
['公式', item.formula],
|
||||
['版本', `v${item.version}`],
|
||||
['预热', item.warmup_bars > 1 ? `${item.warmup_bars} 个交易日` : '无需滚动窗口'],
|
||||
['点时数据', item.pit ? '是（按公告日口径，不泄露未公告数据）' : '否'],
|
||||
['适用资产', item.asset_types.join(' / ')],
|
||||
['跨标的可比', item.scale_free ? '是（可直接截面排序）' : '否（原值受价格尺度影响）'],
|
||||
['依赖列', item.dependencies.length > 0 ? item.dependencies.join(', ') : '已物化列'],
|
||||
['状态', item.stability === 'stable' ? '稳定' : item.stability],
|
||||
]
|
||||
return (
|
||||
<Modal
|
||||
onClose={onClose}
|
||||
labelledBy="factor-detail-title"
|
||||
panelClassName="flex max-h-[86vh] w-[92vw] max-w-lg flex-col overflow-hidden border border-border bg-surface shadow-xl rounded-card"
|
||||
>
|
||||
<div className="flex items-start justify-between gap-3 border-b border-border px-4 py-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<FlaskConical className="h-4 w-4 shrink-0 text-accent" />
|
||||
<span id="factor-detail-title" className="truncate text-sm font-semibold text-foreground">{item.label}</span>
|
||||
<span className={`inline-flex shrink-0 rounded-btn px-1.5 py-0.5 text-[10px] font-medium ${KIND_META[item.kind].cls}`}>
|
||||
{KIND_META[item.kind].label}
|
||||
</span>
|
||||
</div>
|
||||
<div className="mt-1 font-mono text-[11px] text-muted">{item.id}</div>
|
||||
</div>
|
||||
<button
|
||||
type="button"
|
||||
onClick={onClose}
|
||||
className="inline-flex h-8 w-8 shrink-0 items-center justify-center text-muted transition-colors hover:text-foreground"
|
||||
aria-label="关闭"
|
||||
>
|
||||
<X className="h-4 w-4" />
|
||||
</button>
|
||||
</div>
|
||||
<div className="min-h-0 flex-1 overflow-y-auto px-4 py-3">
|
||||
<dl className="space-y-2">
|
||||
{rows.map(([key, value]) => (
|
||||
<div key={key} className="flex items-start gap-3 text-xs">
|
||||
<dt className="w-20 shrink-0 text-muted">{key}</dt>
|
||||
<dd className="min-w-0 flex-1 break-words text-secondary">
|
||||
{key === '分组' && isDynamic ? (
|
||||
<span className="flex items-center gap-1.5" title="自定义/复合因子可修改分组; 内置因子分组固定">
|
||||
<input
|
||||
type="text"
|
||||
value={groupDraft}
|
||||
maxLength={24}
|
||||
onChange={event => setGroupDraft(event.target.value)}
|
||||
className="h-6 w-32 rounded-input border border-border bg-base/60 px-1.5 text-xs text-foreground focus:border-accent focus:outline-none"
|
||||
aria-label="修改分组"
|
||||
/>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setGroup.mutate()}
|
||||
disabled={setGroup.isPending || !groupDraft.trim() || groupDraft.trim() === item.group}
|
||||
className="rounded-btn border border-border px-1.5 py-0.5 text-[10px] text-secondary transition-colors hover:border-accent/40 hover:text-accent disabled:opacity-40"
|
||||
>
|
||||
保存
|
||||
</button>
|
||||
</span>
|
||||
) : value}
|
||||
</dd>
|
||||
</div>
|
||||
))}
|
||||
</dl>
|
||||
<div className="mt-4 rounded-btn border border-border bg-base/40 px-3 py-2 text-[11px] leading-relaxed text-muted">
|
||||
方向说明：因子方向不预填，以最近一次检验的 IC 符号为准（IC 为正 = 值大看多）。到「检验」页运行后可查看。
|
||||
</div>
|
||||
</div>
|
||||
<div className="flex flex-wrap items-center gap-2 border-t border-border px-4 py-3">
|
||||
{isDynamic && (
|
||||
<div className="flex flex-wrap items-center gap-1.5">
|
||||
<span className="text-[10px] text-muted" title="生命周期流转">状态：</span>
|
||||
{(['active', 'watch', 'retired', 'draft'] as const).map(status => (
|
||||
<button
|
||||
key={status}
|
||||
type="button"
|
||||
onClick={() => setStatus.mutate(status)}
|
||||
disabled={setStatus.isPending}
|
||||
className="rounded-btn border border-border px-2 py-0.5 text-[10px] text-secondary transition-colors hover:border-accent/40 hover:text-accent disabled:opacity-40"
|
||||
>
|
||||
{status === 'active' ? '激活' : status === 'watch' ? '观察' : status === 'retired' ? '退役' : '回草稿'}
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
)}
|
||||
{item.kind === 'custom' && onEdit && (
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => { onClose(); onEdit(item.id) }}
|
||||
className="inline-flex items-center gap-1 rounded-btn border border-accent/40 px-2.5 py-1.5 text-[11px] font-medium text-accent transition-colors hover:bg-accent/10"
|
||||
title="在编辑器中打开该公式, 修改后保存为新版本"
|
||||
>
|
||||
<PenLine className="h-3 w-3" />
|
||||
编辑公式
|
||||
</button>
|
||||
)}
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setGenerateOpen(true)}
|
||||
className="inline-flex items-center gap-1 rounded-btn border border-accent/40 px-2.5 py-1.5 text-[11px] font-medium text-accent transition-colors hover:bg-accent/10"
|
||||
title="按该因子生成单因子排名策略, 保存到自定义策略后可直接回测"
|
||||
>
|
||||
<Sparkles className="h-3 w-3" />
|
||||
生成策略
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => { onClose(); onInspect(item.id) }}
|
||||
className="ml-auto inline-flex shrink-0 items-center gap-1 rounded-btn bg-accent px-2.5 py-1.5 text-[11px] font-medium text-white transition-colors hover:bg-accent/90"
|
||||
>
|
||||
检验此因子 <ArrowRight className="h-3 w-3" />
|
||||
</button>
|
||||
{isDynamic && (
|
||||
confirmDelete ? (
|
||||
<>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => removeFactor.mutate(false)}
|
||||
disabled={removeFactor.isPending}
|
||||
className="inline-flex shrink-0 items-center gap-1 rounded-btn bg-danger px-2.5 py-1.5 text-[11px] font-medium text-white transition-colors hover:bg-danger/90 disabled:opacity-40"
|
||||
>
|
||||
{removeFactor.isPending ? '删除中…' : '确认删除'}
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => { setConfirmDelete(false); setBlockedRefs(null) }}
|
||||
className="inline-flex shrink-0 items-center rounded-btn border border-border px-2.5 py-1.5 text-[11px] text-secondary transition-colors hover:text-foreground"
|
||||
>
|
||||
取消
|
||||
</button>
|
||||
</>
|
||||
) : (
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setConfirmDelete(true)}
|
||||
className="inline-flex shrink-0 items-center gap-1 rounded-btn border border-danger/40 px-2.5 py-1.5 text-[11px] font-medium text-danger transition-colors hover:bg-danger/10"
|
||||
title="删除该自定义/复合因子; 被策略或复合因子引用时需强制删除"
|
||||
>
|
||||
<Trash2 className="h-3 w-3" />
|
||||
删除
|
||||
</button>
|
||||
)
|
||||
)}
|
||||
{blockedRefs && (
|
||||
<div className="w-full rounded-btn border border-danger/40 bg-danger/10 px-3 py-2 text-[11px] leading-relaxed text-danger">
|
||||
<div className="font-medium">该因子仍被以下对象引用，删除后相关策略/复合因子将无法计算：</div>
|
||||
<div className="mt-1 font-mono text-[10px] break-all text-danger/90">{blockedRefs.join('、')}</div>
|
||||
<div className="mt-2 flex items-center gap-2">
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => removeFactor.mutate(true)}
|
||||
disabled={removeFactor.isPending}
|
||||
className="rounded-btn bg-danger px-2.5 py-1 text-[11px] font-medium text-white transition-colors hover:bg-danger/90 disabled:opacity-40"
|
||||
>
|
||||
{removeFactor.isPending ? '删除中…' : '仍要强制删除'}
|
||||
</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => setBlockedRefs(null)}
|
||||
className="text-[11px] text-secondary underline-offset-2 transition-colors hover:text-foreground hover:underline"
|
||||
>
|
||||
收起
|
||||
</button>
|
||||
</div>
|
||||
</div>
|
||||
)}
|
||||
</div>
|
||||
{generateOpen && (
|
||||
<GenerateFactorStrategyDialog item={item} onClose={() => setGenerateOpen(false)} />
|
||||
)}
|
||||
</Modal>
|
||||
)
|
||||
}
|
||||
@@ -0,0 +1,152 @@
|
||||
import { useState } from 'react'
|
||||
import { useMutation, useQueryClient } from '@tanstack/react-query'
|
||||
import { Sparkles } from 'lucide-react'
|
||||
import { Modal } from '@/components/Modal'
|
||||
import { toast } from '@/components/Toast'
|
||||
import { api } from '@/lib/api'
|
||||
|
||||
const INPUT_CLS = 'h-8 w-full rounded-input border border-border bg-surface px-2.5 text-xs text-foreground focus:border-accent focus:outline-none'
|
||||
|
||||
/** 渲染与挖掘发布同构的单因子排名策略源码 (FactorRankResearchMatrixStrategy)。 */
|
||||
function renderStrategyCode(strategyId: string, name: string, factorId: string, direction: 'high' | 'low', description: string): string {
|
||||
const assetTypes = '["stock"]'
|
||||
return `"""${name}: 由因子库一键生成的单因子排名策略。"""
|
||||
from app.strategy.builtin.factor_rank_research import FactorRankResearchMatrixStrategy
|
||||
|
||||
META = {
|
||||
"id": ${JSON.stringify(strategyId)},
|
||||
"name": ${JSON.stringify(name)},
|
||||
"description": ${JSON.stringify(description)},
|
||||
"tags": ["factor", "auto-generated"],
|
||||
"asset_types": ${assetTypes},
|
||||
"timeframes": ["1d"],
|
||||
"params": [
|
||||
{"id": "entry_score", "label": "入选评分下限", "type": "float", "default": 70.0, "min": 0.0, "max": 100.0, "step": 5.0},
|
||||
{"id": "exit_score", "label": "离场评分上限", "type": "float", "default": 40.0, "min": 0.0, "max": 100.0, "step": 5.0},
|
||||
{"id": "top_rank", "label": "每日入选上限", "type": "int", "default": 20, "min": 1, "max": 100, "step": 1},
|
||||
],
|
||||
"scoring": {},
|
||||
"order_by": "score",
|
||||
"descending": True,
|
||||
"limit": 100,
|
||||
}
|
||||
|
||||
EXECUTION_BACKEND = "matrix_native"
|
||||
ENTRY_SIGNALS = ["signal_factor_rank_entry"]
|
||||
EXIT_SIGNALS = ["signal_factor_rank_exit"]
|
||||
STOP_LOSS = -0.08
|
||||
MAX_HOLD_DAYS = 30
|
||||
|
||||
SCORING = {${JSON.stringify(factorId)}: 1.0}
|
||||
DIRECTIONS = {${JSON.stringify(factorId)}: ${JSON.stringify(direction)}}
|
||||
MATRIX_STRATEGY = FactorRankResearchMatrixStrategy(SCORING, DIRECTIONS)
|
||||
`
|
||||
}
|
||||
|
||||
/** 因子库 → 一键生成完整可回测策略 (单因子排名, 参数化入场/离场评分与每日上限)。 */
|
||||
export function GenerateFactorStrategyDialog({
|
||||
item, onClose,
|
||||
}: {
|
||||
item: { id: string; label: string; group: string; direction?: 'high' | 'low' | 'none' }
|
||||
onClose: () => void
|
||||
}) {
|
||||
const queryClient = useQueryClient()
|
||||
const [name, setName] = useState(`${item.label}排名策略`)
|
||||
const [direction, setDirection] = useState<'high' | 'low'>(item.direction === 'low' ? 'low' : 'high')
|
||||
|
||||
const generate = useMutation({
|
||||
mutationFn: async () => {
|
||||
const baseId = `custom_factor_${item.id}`.replace(/[^a-z0-9_]/g, '').slice(0, 60)
|
||||
const description = `Generated from factor ${item.id} (${item.label})`
|
||||
const attempts = [baseId, `${baseId}_2`, `${baseId}_3`]
|
||||
let lastError: Error | null = null
|
||||
for (const strategyId of attempts) {
|
||||
try {
|
||||
return await api.strategySaveCodeV2({
|
||||
strategy_id: strategyId,
|
||||
code: renderStrategyCode(strategyId, name.trim(), item.id, direction, description),
|
||||
target_source: 'custom',
|
||||
mode: 'create',
|
||||
name: name.trim(),
|
||||
description,
|
||||
})
|
||||
} catch (error) {
|
||||
lastError = error as Error
|
||||
// ID 冲突时换后缀重试, 其余错误直接抛出
|
||||
if (!/已存在|already exists/i.test((error as Error).message)) throw error
|
||||
}
|
||||
}
|
||||
throw lastError ?? new Error('生成失败')
|
||||
},
|
||||
onSuccess: result => {
|
||||
toast(`已生成策略 ${result.strategy_id}，可到回测页运行`, 'success')
|
||||
void queryClient.invalidateQueries()
|
||||
onClose()
|
||||
},
|
||||
onError: (error: Error) => toast(`生成失败 · ${error.message}`, 'error'),
|
||||
})
|
||||
|
||||
return (
|
||||
<Modal
|
||||
onClose={onClose}
|
||||
labelledBy="generate-factor-strategy-title"
|
||||
panelClassName="flex max-h-[86vh] w-[92vw] max-w-md flex-col overflow-hidden border border-border bg-surface shadow-xl rounded-card"
|
||||
>
|
||||
<div className="flex items-start justify-between gap-3 border-b border-border px-4 py-3">
|
||||
<div className="min-w-0">
|
||||
<div className="flex items-center gap-2">
|
||||
<Sparkles className="h-4 w-4 shrink-0 text-accent" />
|
||||
<span id="generate-factor-strategy-title" className="text-sm font-semibold text-foreground">因子生成策略</span>
|
||||
</div>
|
||||
<p className="mt-1 font-mono text-[11px] text-muted">{item.id}</p>
|
||||
</div>
|
||||
<span className="shrink-0 rounded-btn bg-elevated px-1.5 py-0.5 text-[10px] text-secondary">{item.label}</span>
|
||||
</div>
|
||||
|
||||
<div className="min-h-0 flex-1 space-y-3 overflow-y-auto px-4 py-3">
|
||||
<div className="rounded-btn border border-border bg-base/40 px-3 py-2 text-[11px] leading-relaxed text-secondary">
|
||||
生成一个单因子排名策略：按因子值全市场排名打分，评分高于「入选评分下限」入场、跌破「离场评分上限」出场。
|
||||
入场/离场评分、每日入选上限都是策略参数，回测时可直接调。
|
||||
</div>
|
||||
<label className="block">
|
||||
<span className="mb-1 block text-[10px] text-muted">策略名称</span>
|
||||
<input type="text" value={name} maxLength={24} onChange={event => setName(event.target.value)} className={INPUT_CLS} />
|
||||
</label>
|
||||
<div>
|
||||
<span className="mb-1 block text-[10px] text-muted">因子方向（打分方向）</span>
|
||||
<div className="flex gap-1.5">
|
||||
{([['high', '值大加分'], ['low', '值小加分']] as const).map(([value, label]) => (
|
||||
<button
|
||||
key={value}
|
||||
type="button"
|
||||
onClick={() => setDirection(value)}
|
||||
className={`rounded-btn border px-2.5 py-1.5 text-[11px] transition-colors ${direction === value
|
||||
? 'border-accent/50 bg-accent/10 text-accent'
|
||||
: 'border-border bg-base text-muted hover:border-accent/40'}`}
|
||||
>
|
||||
{label}
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
<p className="mt-1 text-[10px] text-muted">拿不准就先按默认生成，回测对比两个方向再定。</p>
|
||||
</div>
|
||||
<p className="text-[10px] leading-relaxed text-muted">
|
||||
生成后：默认止损 -8%、最大持仓 30 个交易日、评分口径与检验/挖掘同一条计算管线（matrix_native）。策略文件保存在自定义策略目录，可在策略页查看。
|
||||
</p>
|
||||
</div>
|
||||
|
||||
<div className="flex items-center justify-end gap-2 border-t border-border px-4 py-3">
|
||||
<button type="button" onClick={onClose} className="rounded-btn bg-elevated px-3 py-1.5 text-xs text-secondary transition-colors hover:text-foreground">取消</button>
|
||||
<button
|
||||
type="button"
|
||||
onClick={() => generate.mutate()}
|
||||
disabled={generate.isPending || !name.trim()}
|
||||
className="inline-flex items-center gap-1.5 rounded-btn bg-accent px-3 py-1.5 text-xs font-medium text-white transition-colors hover:bg-accent/90 disabled:cursor-not-allowed disabled:opacity-50"
|
||||
>
|
||||
<Sparkles className="h-3.5 w-3.5" />
|
||||
{generate.isPending ? '生成中…' : '生成策略'}
|
||||
</button>
|
||||
</div>
|
||||
</Modal>
|
||||
)
|
||||
}
|
||||
@@ -30,22 +30,25 @@ interface NavEntry {
|
||||
visible: boolean
|
||||
}
|
||||
|
||||
// 与 Layout 侧边栏默认顺序保持一致 (nav_order 未保存时的默认展示顺序)
|
||||
const BUILTIN_PAGES: NavEntry[] = [
|
||||
{ id: '/', label: '看板', type: 'builtin', visible: true },
|
||||
{ id: '/watchlist', label: '自选', type: 'builtin', visible: true },
|
||||
{ id: '/screener', label: '策略', type: 'builtin', visible: true },
|
||||
{ id: '/factors', label: '因子', type: 'builtin', visible: true },
|
||||
{ id: '/backtest', label: '回测', type: 'builtin', visible: true },
|
||||
{ id: '/mining', label: '挖掘', type: 'builtin', visible: true },
|
||||
{ id: '/stock-analysis', label: '个股分析', type: 'builtin', visible: true },
|
||||
{ id: '/limit-ladder', label: '连板梯队', type: 'builtin', visible: true },
|
||||
{ id: '/indices', label: '指数', type: 'builtin', visible: true },
|
||||
{ id: '/concept-analysis', label: '概念分析', type: 'builtin', visible: true },
|
||||
{ id: '/industry-analysis', label: '行业分析', type: 'builtin', visible: true },
|
||||
{ id: '/stock-analysis', label: '个股分析', type: 'builtin', visible: true },
|
||||
{ id: '/regime', label: '市场环境', type: 'builtin', visible: true },
|
||||
{ id: '/abnormal', label: '异动监控', type: 'builtin', visible: true },
|
||||
{ id: '/review', label: '复盘', type: 'builtin', visible: true },
|
||||
{ id: '/financials', label: '财务分析', type: 'builtin', visible: true },
|
||||
{ id: '/monitor', label: '监控中心', type: 'builtin', visible: true },
|
||||
{ id: '/regime', label: '市场环境', type: 'builtin', visible: true },
|
||||
{ id: '/abnormal', label: '异动监控', type: 'builtin', visible: true },
|
||||
{ id: '/lots', label: '持仓提醒', type: 'builtin', visible: true },
|
||||
{ id: '/signals', label: '信号库', type: 'builtin', visible: true },
|
||||
{ id: '/review', label: '复盘', type: 'builtin', visible: true },
|
||||
{ id: '/indices', label: '指数', type: 'builtin', visible: true },
|
||||
{ id: '/data', label: '数据', type: 'builtin', visible: true },
|
||||
]
|
||||
|
||||
|
||||
+14
-3
@@ -1,5 +1,5 @@
|
||||
import { lazy } from 'react'
|
||||
import { createBrowserRouter, Navigate } from 'react-router-dom'
|
||||
import { createBrowserRouter, Navigate, useSearchParams } from 'react-router-dom'
|
||||
import { Layout } from './components/Layout'
|
||||
import { Onboarding } from './pages/Onboarding'
|
||||
import { Auth } from './pages/Auth'
|
||||
@@ -18,7 +18,7 @@ import {
|
||||
const Watchlist = lazy(() => import('./pages/Watchlist').then(m => ({ default: m.Watchlist })))
|
||||
const Screener = lazy(() => import('./pages/Screener').then(m => ({ default: m.Screener })))
|
||||
const Backtest = lazy(() => import('./pages/Backtest').then(m => ({ default: m.Backtest })))
|
||||
const Mining = lazy(() => import('./pages/Mining').then(m => ({ default: m.Mining })))
|
||||
const Factors = lazy(() => import('./pages/Factors').then(m => ({ default: m.Factors })))
|
||||
const Financials = lazy(() => import('./pages/Financials').then(m => ({ default: m.Financials })))
|
||||
const Data = lazy(() => import('./pages/Data').then(m => ({ default: m.Data })))
|
||||
const Monitor = lazy(() => import('./pages/Monitor').then(m => ({ default: m.Monitor })))
|
||||
@@ -28,6 +28,7 @@ const AnalysisDetail = lazy(() => import('./pages/AnalysisDetail').then(m => ({
|
||||
const ConceptAnalysis = lazy(() => import('./pages/ConceptAnalysis').then(m => ({ default: m.ConceptAnalysis })))
|
||||
const IndustryAnalysis = lazy(() => import('./pages/IndustryAnalysis').then(m => ({ default: m.IndustryAnalysis })))
|
||||
const StockAnalysis = lazy(() => import('./pages/StockAnalysis').then(m => ({ default: m.StockAnalysis })))
|
||||
const Signals = lazy(() => import('./pages/Signals').then(m => ({ default: m.Signals })))
|
||||
const Review = lazy(() => import('./pages/Review').then(m => ({ default: m.Review })))
|
||||
const LimitUpLadder = lazy(() => import('./pages/LimitUpLadder').then(m => ({ default: m.LimitUpLadder })))
|
||||
const Indices = lazy(() => import('./pages/Indices').then(m => ({ default: m.Indices })))
|
||||
@@ -51,6 +52,7 @@ const CORE_ROUTE_PATHS = new Set([
|
||||
'/watchlist',
|
||||
'/screener',
|
||||
'/backtest',
|
||||
'/factors',
|
||||
'/mining',
|
||||
'/financials',
|
||||
'/data',
|
||||
@@ -74,6 +76,13 @@ if (frontendExtensionErrors.length > 0) {
|
||||
console.error('部分前端扩展加载失败', frontendExtensionErrors)
|
||||
}
|
||||
|
||||
// 旧链接兼容: 挖掘已并入因子页 (/factors?tab=mining), 保留 run/candidate 等参数重定向
|
||||
function MiningRedirect() {
|
||||
const [searchParams] = useSearchParams()
|
||||
const search = searchParams.toString()
|
||||
return <Navigate to={`/factors?tab=mining${search ? `&${search}` : ''}`} replace />
|
||||
}
|
||||
|
||||
// 首次使用守卫 —— 未完成向导则重定向到 /onboarding
|
||||
// 只挂在根路由上;/onboarding 本身不被守卫,避免循环重定向。
|
||||
// settings 由 Layout 预取,守卫判定不产生额外请求。
|
||||
@@ -124,11 +133,13 @@ export const router = createBrowserRouter([
|
||||
{ path: 'watchlist', element: <Watchlist /> },
|
||||
{ path: 'screener', element: <Screener /> },
|
||||
{ path: 'backtest', element: <Backtest /> },
|
||||
{ path: 'mining', element: <Mining /> },
|
||||
{ path: 'factors', element: <Factors /> },
|
||||
{ path: 'mining', element: <MiningRedirect /> },
|
||||
{ path: 'financials', element: <Financials /> },
|
||||
{ path: 'data', element: <Data /> },
|
||||
{ path: 'monitor', element: <Monitor /> },
|
||||
{ path: 'lots', element: <Lots /> },
|
||||
{ path: 'signals', element: <Signals /> },
|
||||
{ path: 'limit-ladder', element: <LimitUpLadder /> },
|
||||
{ path: 'indices', element: <Indices /> },
|
||||
{ path: 'regime', element: <Regime /> },
|
||||
|
||||
@@ -448,9 +448,12 @@ data/strategies/ai/
|
||||
- 总收益与年化收益。
|
||||
- 夏普比率与索提诺比率。
|
||||
- 最大回撤。
|
||||
- 胜率。
|
||||
- 胜率与盈亏比（平均盈利单收益 ÷ 平均亏损单收益，≥1 显示红色、<1 绿色；判断期望需与胜率一起看）。
|
||||
- 蒙卡回撤（中位 / 95% 边界，把已完成的交易按不同顺序重放 1000 次，估计“仅因成交顺序运气，回撤能有多坏”；中位值对照实际回撤可判断序列运气，资金管理按 95% 边界准备）。
|
||||
- 交易记录和持仓变化。
|
||||
|
||||
若策略评分引用了因子，结果区还会出现「因子归因」标签页：对比盈利单与亏损单在入场信号日的因子均值。胜单均值明显高于败单，说明该因子在本轮交易中贡献了正筛选力；反之在拖后腿。因子名显示为「英文 id + 中文标签」，不同因子量纲不同，只看相对差异。
|
||||
|
||||
历史策略结果与回测交易记录不是同一口径：策略结果表示某个交易日“哪些股票被选中”，回测记录还会叠加成交时序、资金、持仓上限、T+1、费用、止损和卖出规则。因此两者数量或日期不完全一致并不代表系统使用了两套选股逻辑。
|
||||
|
||||
### 9.3 参数优化 Beta
|
||||
|
||||
Reference in New Issue
Block a user