mirror of
https://ghfast.top/https://github.com/aeroxw/tick-stock-panel.git
synced 2026-09-12 15:34:16 +08:00
kevin9327
2266ecc1d8
_compute_rotation_signals 把缺失日的 (999, 0.0) 占位一律追加到序列末尾, 而下游按 ranks[0]=最早日 / ranks[-1]=最新日 解读。概念在部分日期缺席时 (build_rps_rotation 已过滤掉当日无有效 avg_pct 的成员) 实际排名被压到左端、 占位落到右端, 只在最近几日上榜的新晋概念因此被判成退潮。 改为按日期归位补齐, 并补充 tests/test_concept_rotation_signals.py 覆盖 早期缺席 / 最近缺席两个方向。
390 lines
16 KiB
Python
390 lines
16 KiB
Python
"""AI 概念轮动分析 — 从概念涨幅排名矩阵提炼主线/新晋/退潮信号。
|
||
|
||
数据来源:
|
||
- rps_rotation.build_rps_rotation: 概念涨幅排名矩阵 (N 日 × ~387 概念)
|
||
- market_overview_builder.build_market_overview: 大盘背景 (指数/情绪/涨停)
|
||
|
||
架构 (复刻 market_recap):
|
||
预计算轮动信号 → 拼装 prompt → stream_ai_text 流式调用 → NDJSON 协议输出
|
||
协议事件: meta(摘要) / delta(文本片段) / error / done
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import json
|
||
import logging
|
||
import math
|
||
from collections.abc import AsyncIterator
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
|
||
def _dim_label(kind: str) -> str:
|
||
"""维度中文标签: concept→概念, industry→行业。供 prompt/文案动态化。"""
|
||
return "行业" if kind == "industry" else "概念"
|
||
|
||
|
||
# ================================================================
|
||
# System Prompt — 客观轮动分析 + 固定章节模板
|
||
# ================================================================
|
||
|
||
def _build_system_prompt(kind: str = "concept") -> str:
|
||
"""构建 system prompt(按维度动态替换"概念/行业"措辞)。"""
|
||
dim = _dim_label(kind)
|
||
return f"""你是一位专注 A 股题材轮动的研究分析师,拥有 12 年一线研究经验,擅长从{dim}板块的**涨幅排名矩阵**中客观识别主力资金脉络,区分机构主导的持续性主线与游资驱动的脉冲式轮动,产出一份**客观、中立、不包含任何买卖或操作建议**的轮动分析报告。
|
||
|
||
## 核心红线(务必遵守)
|
||
|
||
- **绝对不输出**"跟踪/规避/追高/低吸/观望/操作建议"等任何交易指令或倾向性措辞
|
||
- 你的角色是**客观陈述**各{dim}的轮动特征、资金属性(机构 vs 游资)、持续性特征
|
||
- 换成"一个中立财经记者能不能写出来"——能写就保留,不能写就删除
|
||
|
||
## 输出规范
|
||
|
||
用 **Markdown** 格式输出,严格遵循以下结构。不要输出任何 JSON 或代码块,直接输出 Markdown 正文。
|
||
|
||
### 1. 🎯 主线研判(2-3 句)
|
||
点名当前最核心的 1-2 条主线题材(连续多日霸榜的强势{dim}),用一句话概括其逻辑(政策/产业/业绩/事件驱动),并客观判断是**主升期/加速期/扩散期/见顶期**。结尾用【主线强度:强 / 中 / 弱】客观定性。
|
||
|
||
### 2. 🆕 新晋强势
|
||
列出排名快速跃升的{dim}(从榜单中后段冲进前列的),逐个给出:
|
||
- {dim}名 + 近 N 日排名变化(如 `45→20→8`)
|
||
- 涨幅加速度(连日递增 = 趋势加强)
|
||
- 可能的驱动逻辑(从板块属性推断,不要编造具体消息)
|
||
- 客观判断是**主力切入**还是**消息脉冲**
|
||
|
||
### 3. 📉 退潮预警
|
||
列出从高位明显滑落的{dim}(连续排名下滑或涨幅骤降),逐个给出:
|
||
- {dim}名 + 排名下滑轨迹
|
||
- 退潮性质(高位分歧/资金撤离/补跌)
|
||
- 是否扩散风险
|
||
|
||
### 4. 🏛️ 机构主线 vs 🎰 游资轮动
|
||
基于排名稳定性客观区分两类资金行为:
|
||
- **机构主线**:排名标准差小、长期稳居前列的{dim} → 持续性特征描述
|
||
- **游资轮动**:排名剧烈波动、脉冲式冲高的{dim} → 短线波动特征描述
|
||
客观给出当前市场**整体轮动节奏**(快轮动/慢轮动/主线聚焦)的判断。
|
||
|
||
### 5. 🌐 结合大盘
|
||
结合提供的大盘数据(指数涨跌/情绪/涨停数),客观判断:
|
||
- 当前大盘环境对题材轮动是助力还是阻力
|
||
- 情绪温度与轮动节奏的匹配度(如情绪冰点但题材活跃 = 抱团;情绪火热但轮动快 = 末段)
|
||
|
||
### 6. 📌 后续观察清单
|
||
- **持续性强(客观特征)**:排名标准差小、多日稳居前列的{dim}(列出名称+排名数据)
|
||
- **波动性大(客观特征)**:排名剧烈跳动的{dim}(列出名称+排名数据)
|
||
- 客观描述各{dim}的轮动特征,供读者自行观察
|
||
- **不输出**"跟踪/规避/追高/低吸/观望"等操作指令;可客观描述结构变化信号(如"主线{dim}连续 2 日跌出前 10,主线强度可能减弱")
|
||
|
||
### 7. ⚠️ 风险提示
|
||
列出需要客观关注的风险(如主线断层、情绪与轮动背离、成交萎缩)。末尾附一行:
|
||
"> ⚠️ 本内容由 AI 基于公开行情数据生成,仅客观陈述轮动特征,不构成任何投资建议或买卖指令。交易有风险,入市需谨慎。"
|
||
|
||
## 分析准则(务必遵守)
|
||
|
||
0. **只输出结论,不输出思考过程**:禁止复述你的分析步骤。不要写"我先看...""基于上述数据我认为"——直接给结论。
|
||
1. **数据说话**:每个判断引用具体排名/涨幅数值,严禁空泛套话("强势"必须改成"连续 4 日稳居前 5,均涨 +4.2%")。
|
||
2. **客观中立**:数据不支持的结论就直言"信号不足,暂无法判断",不要硬凑。
|
||
3. **区分资金性质**:这是本分析的核心价值——机构 vs 游资的判断必须基于排名稳定性(标准差),不要凭感觉。
|
||
4. **不重复数字**:正文负责解读信号含义,不要照抄罗列已提供的全部原始数据。
|
||
5. **不输出操作指令**:不写"跟踪/规避/追高/低吸/观望"等任何交易指令;客观陈述轮动特征即可。
|
||
6. **客观推断**:若无明确消息,从量价异动客观推断可能逻辑并给结论,不要标注"[推断]"或编造具体新闻。
|
||
|
||
现在请基于下方{dim}轮动数据进行分析。"""
|
||
|
||
|
||
# ================================================================
|
||
# 预计算: 把排名矩阵转成结构化轮动信号
|
||
# ================================================================
|
||
|
||
# 每类信号最多取多少个概念喂给 AI (控制 token)
|
||
_TOP_N = 8
|
||
|
||
|
||
def _compute_rotation_signals(dates: list[str], columns: dict) -> dict:
|
||
"""从概念涨幅排名矩阵计算轮动信号。
|
||
|
||
Args:
|
||
dates: 日期列表 (最新在最前, 与 columns key 一致)
|
||
columns: {日期: [[概念, 涨幅], ...]} 每列各自降序
|
||
|
||
Returns:
|
||
{
|
||
"persistent_leaders": [...], # 连续多日稳居前列 (主线)
|
||
"rising": [...], # 排名快速跃升 (新晋)
|
||
"fading": [...], # 从高位滑落 (退潮)
|
||
"institutional": [...], # 排名稳定 (机构特征)
|
||
"hot_money": [...], # 排名波动大 (游资特征)
|
||
}
|
||
每项含: concept, ranks (按 dates 顺序), pcts, avg_rank, rank_std
|
||
ranks 时间方向: ranks[0] = 最早日, ranks[-1] = 最新日 (已反转, 左老右新)
|
||
"""
|
||
if not dates or not columns:
|
||
return {}
|
||
|
||
# 按时间正序 (左老右新) 处理
|
||
dates_asc = list(reversed(dates))
|
||
|
||
# 收集每个概念在各日期的 (排名, 涨幅)。排名 = 该日在列中的索引 + 1。
|
||
concept_data: dict[str, dict[str, tuple[int, float]]] = {}
|
||
for d in dates_asc:
|
||
col = columns.get(d) or []
|
||
for idx, (name, pct) in enumerate(col):
|
||
concept_data.setdefault(name, {})[d] = (idx + 1, pct)
|
||
|
||
def _stats(ranks_pcts: list[tuple[int, float]]) -> dict:
|
||
ranks = [r for r, _ in ranks_pcts]
|
||
pcts = [p for _, p in ranks_pcts]
|
||
avg = sum(ranks) / len(ranks) if ranks else 0
|
||
var = sum((r - avg) ** 2 for r in ranks) / len(ranks) if ranks else 0
|
||
return {
|
||
"ranks": ranks,
|
||
"pcts": [round(p, 4) for p in pcts],
|
||
"avg_rank": round(avg, 1),
|
||
"rank_std": round(math.sqrt(var), 1),
|
||
}
|
||
|
||
persistent: list[dict] = []
|
||
rising: list[dict] = []
|
||
fading: list[dict] = []
|
||
institutional: list[dict] = []
|
||
hot_money: list[dict] = []
|
||
|
||
for concept, by_date in concept_data.items():
|
||
# 缺失日按日期归位补 (大排名, 0 涨幅) —— 补位必须落在缺席的那一天,
|
||
# 一律追加到末尾会把"只在最近几日上榜"的新晋概念读成退潮。
|
||
rp = [by_date.get(d, (999, 0.0)) for d in dates_asc]
|
||
s = _stats(rp)
|
||
s["concept"] = concept
|
||
|
||
ranks = s["ranks"]
|
||
latest_rank = ranks[-1]
|
||
earliest_rank = ranks[0]
|
||
# 最近 3 日 (不足则全部) 均排名, 判断近期强度
|
||
recent = ranks[-min(3, len(ranks)):]
|
||
recent_avg = sum(recent) / len(recent)
|
||
|
||
# 主线: 近期稳居前 10
|
||
if recent_avg <= 10 and latest_rank <= 10:
|
||
persistent.append(s)
|
||
|
||
# 新晋: 早期排名靠后(>30), 最新冲进前 20, 跃升幅度大
|
||
jump = earliest_rank - latest_rank
|
||
if earliest_rank > 30 and latest_rank <= 20 and jump >= 20:
|
||
rising.append(s)
|
||
|
||
# 退潮: 早期排名靠前(<=10), 最新滑落到 30 外
|
||
drop = latest_rank - earliest_rank
|
||
if earliest_rank <= 10 and latest_rank > 30 and drop >= 20:
|
||
fading.append(s)
|
||
|
||
# 机构: 排名标准差小且平均排名靠前 (稳定强势)
|
||
if s["rank_std"] <= 5 and s["avg_rank"] <= 20:
|
||
institutional.append(s)
|
||
|
||
# 游资: 排名标准差大 (波动剧烈)
|
||
if s["rank_std"] >= 20:
|
||
hot_money.append(s)
|
||
|
||
# 排序: 主线按近期排名升序; 新晋按跃升幅度降序; 退潮按跌幅降序
|
||
persistent.sort(key=lambda x: x["avg_rank"])
|
||
rising.sort(key=lambda x: x["ranks"][0] - x["ranks"][-1], reverse=True)
|
||
fading.sort(key=lambda x: x["ranks"][-1] - x["ranks"][0], reverse=True)
|
||
institutional.sort(key=lambda x: (x["rank_std"], x["avg_rank"]))
|
||
hot_money.sort(key=lambda x: x["rank_std"], reverse=True)
|
||
|
||
return {
|
||
"persistent_leaders": persistent[:_TOP_N],
|
||
"rising": rising[:_TOP_N],
|
||
"fading": fading[:_TOP_N],
|
||
"institutional": institutional[:_TOP_N],
|
||
"hot_money": hot_money[:_TOP_N],
|
||
}
|
||
|
||
|
||
# ================================================================
|
||
# Prompt 构建
|
||
# ================================================================
|
||
|
||
def _fmt_pct(v) -> str:
|
||
if v is None:
|
||
return "—"
|
||
return f"{v*100:+.2f}%"
|
||
|
||
|
||
def _build_market_block(overview: dict) -> str:
|
||
"""大盘背景精简块 (复用 market_overview 已算好的字段)。"""
|
||
indices = overview.get("indices") or []
|
||
emo = overview.get("emotion") or {}
|
||
lim = overview.get("limit") or {}
|
||
amt = overview.get("amount") or {}
|
||
|
||
idx_lines = []
|
||
for idx in indices[:4]:
|
||
name = idx.get("name") or idx.get("symbol") or "?"
|
||
chg = idx.get("change_pct")
|
||
idx_lines.append(f"{name} {_fmt_pct(chg)}")
|
||
idx_str = " / ".join(idx_lines) or "指数缺失"
|
||
|
||
total_amount = (amt.get("total") or 0) / 1e8 # 元 → 亿
|
||
|
||
return (
|
||
f"- 指数: {idx_str}\n"
|
||
f"- 情绪: {emo.get('score', 50)} ({emo.get('label', '—')})\n"
|
||
f"- 涨停/炸板/跌停: {lim.get('limit_up', 0)} / {lim.get('broken', 0)} / {lim.get('limit_down', 0)}"
|
||
f" (最高连板 {lim.get('max_boards', 0)})\n"
|
||
f"- 两市成交额: {total_amount:.0f} 亿元"
|
||
)
|
||
|
||
|
||
def _build_signal_block(title: str, items: list[dict]) -> str:
|
||
"""轮动信号块: 把预计算的概念信号转成紧凑文本。"""
|
||
if not items:
|
||
return f"### {title}\n(本类无明显信号)"
|
||
lines = [f"### {title}"]
|
||
for it in items:
|
||
ranks_str = "→".join(str(r) if r < 999 else "—" for r in it["ranks"])
|
||
avg_pct = sum(it["pcts"]) / len(it["pcts"]) if it["pcts"] else 0
|
||
lines.append(
|
||
f"- {it['concept']}: 排名 {ranks_str} | 均排名 {it['avg_rank']} "
|
||
f"| 排名波动σ {it['rank_std']} | 区间均涨 {_fmt_pct(avg_pct)}"
|
||
)
|
||
return "\n".join(lines)
|
||
|
||
|
||
def _build_user_prompt(signals: dict, overview: dict, days: int, dates: list[str], focus: str, kind: str = "concept") -> str:
|
||
"""组装 user 消息: 大盘背景 + 轮动信号 + focus。"""
|
||
dim = _dim_label(kind)
|
||
dates_asc = list(reversed(dates))
|
||
date_range = f"{dates_asc[0]} ~ {dates_asc[-1]}" if dates_asc else "—"
|
||
|
||
parts = [
|
||
f"# {dim}涨幅轮动数据 (最近 {days} 个交易日: {date_range})",
|
||
"",
|
||
"## 大盘背景",
|
||
_build_market_block(overview),
|
||
"",
|
||
"## 轮动信号 (排名时间方向: 左→右 = 旧→新, 排名越小越强)",
|
||
"",
|
||
_build_signal_block("🎯 主线 (连续霸榜)", signals.get("persistent_leaders", [])),
|
||
"",
|
||
_build_signal_block("🆕 新晋强势 (排名跃升)", signals.get("rising", [])),
|
||
"",
|
||
_build_signal_block("📉 退潮预警 (高位滑落)", signals.get("fading", [])),
|
||
"",
|
||
_build_signal_block("🏛️ 机构特征 (排名稳定)", signals.get("institutional", [])),
|
||
"",
|
||
_build_signal_block("🎰 游资特征 (排名波动大)", signals.get("hot_money", [])),
|
||
]
|
||
|
||
from app.services.ai_provider import build_focus_instruction
|
||
focus_instruction = build_focus_instruction(focus, report_name=f"{dim}轮动分析报告")
|
||
if focus_instruction:
|
||
parts.extend(["", focus_instruction])
|
||
|
||
return "\n".join(parts)
|
||
|
||
|
||
def _build_summary(signals: dict) -> str:
|
||
"""meta 事件的摘要 (前端可立即展示)。"""
|
||
leaders = signals.get("persistent_leaders", [])
|
||
rising = signals.get("rising", [])
|
||
fading = signals.get("fading", [])
|
||
leader_names = "、".join(it["concept"] for it in leaders[:3]) or "暂无明确主线"
|
||
return f"主线: {leader_names} | 新晋 {len(rising)} | 退潮 {len(fading)}"
|
||
|
||
|
||
# ================================================================
|
||
# 流式主入口
|
||
# ================================================================
|
||
|
||
async def analyze_rotation_stream(
|
||
repo,
|
||
days: int = 12,
|
||
focus: str = "",
|
||
quote_service=None,
|
||
depth_service=None,
|
||
kind: str = "concept",
|
||
level: int | None = None,
|
||
) -> AsyncIterator[str]:
|
||
"""流式维度轮动分析(概念或行业): yield 出每个 NDJSON 事件。
|
||
|
||
Args:
|
||
repo: KlineRepository (必填)。
|
||
days: 分析最近 N 个交易日 (7-30)。
|
||
focus: 用户追加的关注点。
|
||
quote_service / depth_service: 可选, 大盘背景装配依赖。
|
||
kind: "concept"(概念) 或 "industry"(行业)。
|
||
level: 行业层级(1/2/3), 仅 kind=industry 有效。
|
||
"""
|
||
from app.services.rps_rotation import build_rps_rotation
|
||
from app.services.market_overview_builder import build_market_overview
|
||
|
||
dim = _dim_label(kind)
|
||
page = "行业分析" if kind == "industry" else "概念分析"
|
||
|
||
# 1. 取轮动矩阵
|
||
rotation = build_rps_rotation(repo, days, kind, level)
|
||
dates = rotation.get("dates") or []
|
||
columns = rotation.get("columns") or {}
|
||
|
||
if not dates or not columns:
|
||
yield json.dumps({
|
||
"type": "error",
|
||
"message": f"暂无{dim}轮动数据,请先在「{page}」页获取{dim}数据源",
|
||
}, ensure_ascii=False)
|
||
return
|
||
|
||
# 2. 预计算轮动信号
|
||
signals = _compute_rotation_signals(dates, columns)
|
||
|
||
# 3. 大盘背景 (失败不阻断, 降级为空)
|
||
try:
|
||
overview = build_market_overview(repo, quote_service, depth_service)
|
||
except Exception as e: # noqa: BLE001
|
||
logger.warning("rotation analyze: 大盘背景获取失败, 降级为空: %s", e)
|
||
overview = {}
|
||
|
||
# 4. meta 事件
|
||
yield json.dumps({
|
||
"type": "meta",
|
||
"days": days,
|
||
"summary": _build_summary(signals),
|
||
}, ensure_ascii=False)
|
||
|
||
# 5. 构建 prompt + 流式调用 LLM
|
||
try:
|
||
from app.services.ai_provider import stream_ai_text, ai_configured
|
||
|
||
if not ai_configured():
|
||
yield json.dumps({
|
||
"type": "error",
|
||
"message": "AI 未配置,请在「设置」页填写 API Key 与接口地址",
|
||
}, ensure_ascii=False)
|
||
return
|
||
|
||
user_prompt = _build_user_prompt(signals, overview, days, dates, focus, kind)
|
||
got_content = False
|
||
async for delta in stream_ai_text(
|
||
[
|
||
{"role": "system", "content": _build_system_prompt(kind)},
|
||
{"role": "user", "content": user_prompt},
|
||
],
|
||
temperature=0.5,
|
||
# 不限制输出(推理模型思考 token 计入预算, 见 ai_provider.stream_ai_text)
|
||
max_tokens=None,
|
||
prefer_final_answer=True,
|
||
):
|
||
got_content = True
|
||
yield json.dumps({"type": "delta", "content": delta}, ensure_ascii=False)
|
||
|
||
except Exception as e: # noqa: BLE001
|
||
logger.exception("AI %s rotation analyze failed: %s", kind, e)
|
||
yield json.dumps({"type": "error", "message": f"AI 轮动分析失败: {e}"}, ensure_ascii=False)
|
||
return
|
||
|
||
if not got_content:
|
||
logger.warning("AI %s rotation analyze ended with empty content", kind)
|
||
yield json.dumps({"type": "error", "message": "AI 未返回正文(输出被截断), 请重试"}, ensure_ascii=False)
|
||
return
|
||
yield json.dumps({"type": "done"}, ensure_ascii=False)
|