mirror of
https://ghfast.top/https://github.com/aeroxw/tick-stock-panel.git
synced 2026-09-12 21:24:16 +08:00
kevin9327andClaude Opus 4.8
d92ef17851
问题: /api/backtest/run 设置 max_hold_days 时, 强制平仓完全不生效, 持仓不会 按最大持有天数退出, 导致交易记录、持仓周期和收益统计失真。 根因(services/backtest.py 的 max_hold 内联逻辑, 两处缺陷叠加): 1. `exits_idx.iloc[end_i][col] = True` 是链式索引: iloc[end_i] 先取出一行副本, 再对副本赋值, 在 pandas Copy-on-Write 语义下写入不会落到原矩阵(pandas 3.x 直接抛 ChainedAssignmentError)。强制退出单元格始终为 False。 2. `exits_idx = entries.copy()` 以入场矩阵起步, 使强制退出矩阵天然带上所有入场位; 即便修好第 1 点, 也会在入场当日就强制平仓, 而非 max_hold_days 之后。 修复: 抽出纯函数 _build_max_hold_exits(entries, max_hold_days), 从全 False 起步, 用单步定位 iloc[row, col_loc] 写入入场后第 max_hold_days 个交易日的强制退出位; run() 中改为与用户 exits 做 OR, 保留原有信号退出。抽出后该逻辑不依赖 vectorbt, 可独立回归(run() 整体仍需 vectorbt 可选依赖)。 边界: end_i 越界 clamp 到最后一根 K; 入场即最后一根 K 时 end_i==i 不产生退出。 验证: 新增 tests/backtest/test_max_hold_exits.py 覆盖强制退出落位、不误标入场位、 越界 clamp、末根不退出、多入场、多列独立 6 例, 修复后全过; 并以复刻旧内联逻辑的 脚本在 pandas 3.0.5 上确认修复前目标退出单元格恒为 False(ChainedAssignmentError)。 ruff 通过。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
417 lines
15 KiB
Python
417 lines
15 KiB
Python
"""回测服务(§6.7)。
|
||
|
||
包 vectorbt — 全项目唯一一处出现 pandas。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import logging
|
||
import uuid
|
||
from dataclasses import dataclass, field
|
||
from datetime import date
|
||
from typing import Literal
|
||
|
||
import numpy as np
|
||
import pandas as pd
|
||
import polars as pl
|
||
|
||
from app.config import settings
|
||
from app.parquet import scan_enriched_parquet
|
||
from app.tickflow.repository import KlineRepository
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
# vectorbt 是 optional extras(见 pyproject.toml).未装时只有 backtest 不可用,其他功能正常.
|
||
_vbt = None
|
||
_vbt_unavailable_reason: str | None = None
|
||
|
||
|
||
class VectorbtUnavailable(RuntimeError):
|
||
"""vectorbt 未安装 — 提示用户 `uv sync --extra backtest`."""
|
||
|
||
|
||
def _get_vbt():
|
||
global _vbt, _vbt_unavailable_reason
|
||
if _vbt is not None:
|
||
return _vbt
|
||
if _vbt_unavailable_reason is not None:
|
||
raise VectorbtUnavailable(_vbt_unavailable_reason)
|
||
try:
|
||
import vectorbt as vbt
|
||
_vbt = vbt
|
||
return _vbt
|
||
except ImportError as e:
|
||
_vbt_unavailable_reason = (
|
||
"vectorbt 未安装 — 它是回测的可选依赖.macOS Intel 用户先 `brew install cmake` "
|
||
"然后 `uv sync --extra backtest`"
|
||
)
|
||
logger.warning("vectorbt unavailable: %s", e)
|
||
raise VectorbtUnavailable(_vbt_unavailable_reason) from e
|
||
|
||
|
||
def is_available() -> bool:
|
||
"""供 API 层快速检测."""
|
||
try:
|
||
_get_vbt()
|
||
return True
|
||
except VectorbtUnavailable:
|
||
return False
|
||
|
||
|
||
SignalKind = Literal[
|
||
"macd_golden", "macd_dead",
|
||
"ma_golden_5_20", "ma_dead_5_20",
|
||
"ma_golden_20_60",
|
||
"ma20_breakout", "ma20_breakdown",
|
||
"n_day_high", "n_day_low",
|
||
"boll_breakout_upper", "boll_breakdown_lower",
|
||
"volume_surge",
|
||
"rsi_oversold", "rsi_overbought",
|
||
"stop_loss", "trailing_stop", "max_hold",
|
||
]
|
||
|
||
|
||
@dataclass
|
||
class BacktestConfig:
|
||
symbols: list[str]
|
||
start: date
|
||
end: date
|
||
# 买入信号(任一触发即买)
|
||
entries: list[str] = field(default_factory=list)
|
||
# 卖出信号(任一触发即卖)
|
||
exits: list[str] = field(default_factory=list)
|
||
# 其他参数
|
||
stop_loss_pct: float | None = None # 例 -0.05 = -5%
|
||
max_hold_days: int | None = None
|
||
fees_pct: float = 0.0002 # 万二佣金
|
||
slippage_bps: float = 5 # 5 bps
|
||
# 撮合
|
||
matching: Literal["close_t", "open_t+1"] = "close_t"
|
||
rsi_oversold_threshold: float = 30
|
||
rsi_overbought_threshold: float = 70
|
||
asset_type: str = "stock"
|
||
|
||
|
||
@dataclass
|
||
class BacktestResult:
|
||
run_id: str
|
||
config: dict
|
||
stats: dict
|
||
equity_curve: list[dict] # [{date, value}]
|
||
trades: list[dict] # [{symbol, entry_date, exit_date, pnl_pct, ...}]
|
||
per_symbol_stats: list[dict] # 每只股票的统计
|
||
|
||
|
||
# enriched 表里的信号列名映射
|
||
_SIGNAL_COLS: dict[SignalKind, str] = {
|
||
"macd_golden": "signal_macd_golden",
|
||
"macd_dead": "signal_macd_dead",
|
||
"ma_golden_5_20": "signal_ma_golden_5_20",
|
||
"ma_dead_5_20": "signal_ma_dead_5_20",
|
||
"ma_golden_20_60": "signal_ma_golden_20_60",
|
||
"ma20_breakout": "signal_ma20_breakout",
|
||
"ma20_breakdown": "signal_ma20_breakdown",
|
||
"n_day_high": "signal_n_day_high",
|
||
"n_day_low": "signal_n_day_low",
|
||
"boll_breakout_upper": "signal_boll_breakout_upper",
|
||
"boll_breakdown_lower": "signal_boll_breakdown_lower",
|
||
"volume_surge": "signal_volume_surge",
|
||
}
|
||
|
||
|
||
def _build_max_hold_exits(entries: pd.DataFrame, max_hold_days: int) -> pd.DataFrame:
|
||
"""为每个入场信号在 max_hold_days 个交易日后生成一个强制退出信号。
|
||
|
||
返回与 entries 同形状的布尔矩阵, 仅在「入场位之后第 max_hold_days 个交易日」置
|
||
True(不含入场位本身), 供调用方与用户 exits 做 OR。
|
||
|
||
两处易错点(见 #198):
|
||
- 必须从全 False 起步。若用 `entries.copy()` 起步会把入场位当成退出位,
|
||
导致入场当日即被强制平仓。
|
||
- 用单步定位写入 `iloc[row, col_loc]`。链式 `iloc[row][col] = True` 写入的是
|
||
临时行副本, 在 pandas Copy-on-Write 语义下不会落到原矩阵(pandas 3.x 直接报错),
|
||
强制退出信号会静默丢失。
|
||
"""
|
||
out = pd.DataFrame(False, index=entries.index, columns=entries.columns)
|
||
n = len(entries)
|
||
for col in entries.columns:
|
||
col_loc = out.columns.get_loc(col)
|
||
entry_rows = np.where(entries[col].to_numpy())[0]
|
||
for i in entry_rows:
|
||
end_i = min(int(i) + max_hold_days, n - 1)
|
||
if end_i > i:
|
||
out.iloc[end_i, col_loc] = True
|
||
return out
|
||
|
||
|
||
class BacktestService:
|
||
def __init__(self, repo: KlineRepository) -> None:
|
||
self.repo = repo
|
||
|
||
def _load_panel(
|
||
self,
|
||
symbols: list[str],
|
||
start: date,
|
||
end: date,
|
||
asset_type: str = "stock",
|
||
) -> pd.DataFrame:
|
||
"""加载 [date × symbol] 价格面板 — Polars scan_parquet + 即时计算指标。
|
||
|
||
**全项目唯一从 Polars 转 pandas 的边界**(§7.4 / ADR-19)。
|
||
asset_type='etf' 时读 ETF enriched。
|
||
"""
|
||
try:
|
||
from app.tickflow.repository import enriched_dirname
|
||
enriched_glob = str(self.repo.store.data_dir / enriched_dirname(asset_type) / "**" / "*.parquet")
|
||
df = (
|
||
scan_enriched_parquet(enriched_glob)
|
||
.filter(
|
||
(pl.col("symbol").is_in(symbols))
|
||
& (pl.col("date") >= start)
|
||
& (pl.col("date") <= end)
|
||
)
|
||
.sort(["date", "symbol"])
|
||
.collect()
|
||
)
|
||
except Exception as e: # noqa: BLE001
|
||
logger.warning("backtest load failed: %s", e)
|
||
return pd.DataFrame()
|
||
|
||
if df.is_empty():
|
||
return pd.DataFrame()
|
||
|
||
# 即时计算指标 + 信号
|
||
from app.indicators.pipeline import compute_all
|
||
df = compute_all(df)
|
||
|
||
# 选择需要的列
|
||
needed_cols = [
|
||
"date", "symbol", "open", "high", "low", "close", "volume",
|
||
"rsi_14", "signal_macd_golden", "signal_macd_dead",
|
||
"signal_ma_golden_5_20", "signal_ma_dead_5_20",
|
||
"signal_ma_golden_20_60",
|
||
"signal_ma20_breakout", "signal_ma20_breakdown",
|
||
"signal_n_day_high", "signal_n_day_low",
|
||
"signal_boll_breakout_upper", "signal_boll_breakdown_lower",
|
||
"signal_volume_surge",
|
||
]
|
||
existing = [c for c in needed_cols if c in df.columns]
|
||
df = df.select(existing)
|
||
|
||
# to_pandas 边界
|
||
return df.to_pandas(use_pyarrow_extension_array=False)
|
||
|
||
def _build_signal_matrix(
|
||
self,
|
||
panel: pd.DataFrame,
|
||
kinds: list[str],
|
||
config: BacktestConfig,
|
||
) -> pd.DataFrame:
|
||
"""从面板构造 [date × symbol] 的布尔信号矩阵。"""
|
||
if not kinds or panel.empty:
|
||
return pd.DataFrame()
|
||
|
||
# pivot 成 [date × symbol] 形式
|
||
result = None
|
||
for kind in kinds:
|
||
mat = None
|
||
if kind in _SIGNAL_COLS:
|
||
col = _SIGNAL_COLS[kind]
|
||
mat = panel.pivot(index="date", columns="symbol", values=col).fillna(False).astype(bool)
|
||
elif kind == "rsi_oversold":
|
||
mat = (panel.pivot(index="date", columns="symbol", values="rsi_14")
|
||
< config.rsi_oversold_threshold)
|
||
elif kind == "rsi_overbought":
|
||
mat = (panel.pivot(index="date", columns="symbol", values="rsi_14")
|
||
> config.rsi_overbought_threshold)
|
||
# stop_loss / trailing / max_hold 通过 vectorbt 参数处理,不参与信号矩阵
|
||
|
||
if mat is not None:
|
||
result = mat if result is None else (result | mat)
|
||
return result if result is not None else pd.DataFrame()
|
||
|
||
def run(self, config: BacktestConfig) -> BacktestResult:
|
||
vbt = _get_vbt()
|
||
run_id = uuid.uuid4().hex[:10]
|
||
|
||
panel = self._load_panel(config.symbols, config.start, config.end, config.asset_type)
|
||
if panel.empty:
|
||
return BacktestResult(
|
||
run_id=run_id,
|
||
config=_config_to_dict(config),
|
||
stats={"error": "no data"},
|
||
equity_curve=[],
|
||
trades=[],
|
||
per_symbol_stats=[],
|
||
)
|
||
|
||
# 价格面板
|
||
close = panel.pivot(index="date", columns="symbol", values="close")
|
||
|
||
# 信号矩阵
|
||
entries = self._build_signal_matrix(panel, config.entries, config)
|
||
exits = self._build_signal_matrix(panel, config.exits, config)
|
||
|
||
# 对齐 index/columns
|
||
if not entries.empty:
|
||
entries = entries.reindex_like(close).fillna(False).astype(bool)
|
||
else:
|
||
entries = pd.DataFrame(False, index=close.index, columns=close.columns)
|
||
if not exits.empty:
|
||
exits = exits.reindex_like(close).fillna(False).astype(bool)
|
||
else:
|
||
exits = pd.DataFrame(False, index=close.index, columns=close.columns)
|
||
|
||
if not entries.any().any():
|
||
return BacktestResult(
|
||
run_id=run_id,
|
||
config=_config_to_dict(config),
|
||
stats={"error": "no buy signals"},
|
||
equity_curve=[],
|
||
trades=[],
|
||
per_symbol_stats=[],
|
||
)
|
||
|
||
# T+1 适配:vectorbt 默认信号当根 K 撮合
|
||
# close_t 撮合:维持默认
|
||
# open_t+1 撮合:shift 信号 1 根 + 用 open 作为价
|
||
if config.matching == "open_t+1":
|
||
entries = entries.shift(1).fillna(False).astype(bool)
|
||
exits = exits.shift(1).fillna(False).astype(bool)
|
||
price = panel.pivot(index="date", columns="symbol", values="open")
|
||
else:
|
||
price = close
|
||
|
||
# 跑回测
|
||
try:
|
||
pf_kwargs = dict(
|
||
close=close,
|
||
entries=entries,
|
||
exits=exits,
|
||
price=price,
|
||
fees=config.fees_pct,
|
||
slippage=config.slippage_bps / 10000.0,
|
||
freq="1D",
|
||
)
|
||
if config.stop_loss_pct is not None:
|
||
pf_kwargs["sl_stop"] = abs(config.stop_loss_pct)
|
||
if config.max_hold_days is not None:
|
||
# vectorbt 没有内置 max-hold;用时间退出近似:入场后第 max_hold_days
|
||
# 个交易日强制 exit, 与用户 exits 做 OR(保留原有信号退出)。
|
||
forced_exits = _build_max_hold_exits(entries, config.max_hold_days)
|
||
pf_kwargs["exits"] = (exits | forced_exits).astype(bool)
|
||
|
||
pf = vbt.Portfolio.from_signals(**pf_kwargs)
|
||
except Exception as e: # noqa: BLE001
|
||
logger.exception("vectorbt backtest failed")
|
||
return BacktestResult(
|
||
run_id=run_id,
|
||
config=_config_to_dict(config),
|
||
stats={"error": str(e)},
|
||
equity_curve=[],
|
||
trades=[],
|
||
per_symbol_stats=[],
|
||
)
|
||
|
||
# 提取结果
|
||
try:
|
||
stats_series = pf.stats(silence_warnings=True)
|
||
if isinstance(stats_series, pd.DataFrame):
|
||
# 多列时取 agg
|
||
stats_dict = stats_series.mean(numeric_only=True).to_dict()
|
||
else:
|
||
stats_dict = stats_series.to_dict()
|
||
except Exception: # noqa: BLE001
|
||
stats_dict = {}
|
||
|
||
# 净值曲线(组合平均)
|
||
equity = pf.value().mean(axis=1) if isinstance(pf.value(), pd.DataFrame) else pf.value()
|
||
equity_curve = [
|
||
{"date": str(idx.date() if hasattr(idx, "date") else idx), "value": float(v)}
|
||
for idx, v in equity.items() if pd.notna(v)
|
||
]
|
||
|
||
# 交易记录
|
||
try:
|
||
trades_df = pf.trades.records_readable
|
||
trades = trades_df.to_dict(orient="records") if not trades_df.empty else []
|
||
# 字段名美化
|
||
trades = [
|
||
{
|
||
"symbol": t.get("Column", t.get("Symbol", "")),
|
||
"entry_date": str(t.get("Entry Timestamp", t.get("Entry Date", ""))),
|
||
"exit_date": str(t.get("Exit Timestamp", t.get("Exit Date", ""))),
|
||
"entry_price": float(t.get("Avg Entry Price", t.get("Avg. Entry Price", 0))),
|
||
"exit_price": float(t.get("Avg Exit Price", t.get("Avg. Exit Price", 0))),
|
||
"pnl_pct": float(t.get("Return", t.get("PnL %", 0))),
|
||
"duration": str(t.get("Duration", "")),
|
||
}
|
||
for t in trades
|
||
]
|
||
except Exception: # noqa: BLE001
|
||
trades = []
|
||
|
||
# 每标的统计
|
||
per_symbol = []
|
||
try:
|
||
total_ret = pf.total_return()
|
||
if isinstance(total_ret, pd.Series):
|
||
for sym, ret in total_ret.items():
|
||
if pd.notna(ret):
|
||
per_symbol.append({"symbol": sym, "total_return": float(ret)})
|
||
except Exception: # noqa: BLE001
|
||
pass
|
||
|
||
result = BacktestResult(
|
||
run_id=run_id,
|
||
config=_config_to_dict(config),
|
||
stats={k: _json_safe(v) for k, v in stats_dict.items()},
|
||
equity_curve=equity_curve,
|
||
trades=trades,
|
||
per_symbol_stats=per_symbol,
|
||
)
|
||
|
||
# 落盘
|
||
self._persist(result)
|
||
return result
|
||
|
||
def _persist(self, result: BacktestResult) -> None:
|
||
out_dir = settings.data_dir / "backtest_results"
|
||
out_dir.mkdir(parents=True, exist_ok=True)
|
||
# 用 polars 写一份汇总
|
||
summary = pl.DataFrame({
|
||
"run_id": [result.run_id],
|
||
"stats_json": [str(result.stats)],
|
||
"n_trades": [len(result.trades)],
|
||
})
|
||
summary.write_parquet(out_dir / f"run_id={result.run_id}.parquet")
|
||
|
||
def get_result(self, run_id: str) -> BacktestResult | None:
|
||
# Phase 1:只保留近似落盘,完整结果保存在内存的近期 cache 中
|
||
# 简化:重新 run 比缓存复杂结果代价小,暂不实现 get_result
|
||
return None
|
||
|
||
|
||
def _config_to_dict(c: BacktestConfig) -> dict:
|
||
return {
|
||
"symbols": c.symbols,
|
||
"start": str(c.start),
|
||
"end": str(c.end),
|
||
"entries": c.entries,
|
||
"exits": c.exits,
|
||
"stop_loss_pct": c.stop_loss_pct,
|
||
"max_hold_days": c.max_hold_days,
|
||
"fees_pct": c.fees_pct,
|
||
"slippage_bps": c.slippage_bps,
|
||
"matching": c.matching,
|
||
}
|
||
|
||
|
||
def _json_safe(v):
|
||
if isinstance(v, (int, float, str, bool)) or v is None:
|
||
return v
|
||
if isinstance(v, (np.floating, np.integer)):
|
||
return float(v) if not np.isnan(float(v)) else None
|
||
if hasattr(v, "isoformat"):
|
||
return v.isoformat()
|
||
return str(v)
|