233 lines
13 KiB
Python
233 lines
13 KiB
Python
"""AI 分析层：基于中台数据构造上下文，调用大模型生成点评；无 key 时规则降级。"""
|
||
from __future__ import annotations
|
||
|
||
import datetime as dt
|
||
|
||
from sqlalchemy import select, func, desc
|
||
|
||
import llm
|
||
import signals
|
||
import rag
|
||
from db import get_session
|
||
from models import (SectorDaily, FundFlowDaily, SentimentDaily, DragonTiger,
|
||
StockMetric, DailyQuote, Security)
|
||
|
||
DISCLAIMER = "（注：以上为基于数据的程序化分析，不构成投资建议，市场有风险。）"
|
||
|
||
|
||
# ============ 每日复盘点评 ============
|
||
def _daily_context(date=None):
|
||
with get_session() as s:
|
||
d = dt.date.fromisoformat(date) if date else s.execute(select(func.max(SectorDaily.date))).scalar()
|
||
if not d:
|
||
return None
|
||
secs = s.execute(select(SectorDaily).where(SectorDaily.date == d).order_by(SectorDaily.pct.desc())).scalars().all()
|
||
flows = s.execute(select(FundFlowDaily).where(FundFlowDaily.date == d).order_by(FundFlowDaily.net.desc())).scalars().all()
|
||
senti = s.execute(select(SentimentDaily).where(SentimentDaily.date == d)).scalar_one_or_none()
|
||
lhb = s.execute(select(DragonTiger).where(DragonTiger.date == d).order_by(DragonTiger.net.desc()).limit(6)).scalars().all()
|
||
return {
|
||
"date": d, "senti": senti,
|
||
"top": [(x.name, x.pct) for x in secs[:6]],
|
||
"bot": [(x.name, x.pct) for x in secs[-4:]],
|
||
"inflow": [(x.name, x.net) for x in flows[:6]],
|
||
"outflow": [(x.name, x.net) for x in flows[-4:][::-1]],
|
||
"lhb": [(x.name, x.net) for x in lhb],
|
||
}
|
||
|
||
|
||
def review_daily_comment(date=None):
|
||
ctx = _daily_context(date)
|
||
if not ctx:
|
||
return {"ok": False, "msg": "暂无入库数据，请先到数据中台入库"}
|
||
s = ctx["senti"]
|
||
senti_line = (f"上涨{s.up}家/下跌{s.down}家，涨停{s.limit_up}家/跌停{s.limit_down}家" if s else "无情绪数据")
|
||
facts = (
|
||
f"日期：{ctx['date']}\n"
|
||
f"市场情绪：{senti_line}\n"
|
||
f"领涨板块：{', '.join(f'{n}({p:+.2f}%)' for n,p in ctx['top'])}\n"
|
||
f"领跌板块：{', '.join(f'{n}({p:+.2f}%)' for n,p in ctx['bot'])}\n"
|
||
f"主力净流入前列：{', '.join(f'{n}({v:+.1f}亿)' for n,v in ctx['inflow'])}\n"
|
||
f"主力净流出前列：{', '.join(f'{n}({v:+.1f}亿)' for n,v in ctx['outflow'])}\n"
|
||
f"龙虎榜净买额居前：{', '.join(f'{n}({v:+.2f}亿)' for n,v in ctx['lhb'])}\n"
|
||
)
|
||
if llm.enabled():
|
||
try:
|
||
prompt = ("请根据以下当日A股盘面数据，撰写一篇 250 字以内的收盘复盘点评，"
|
||
"包含：① 今日市场情绪与赚钱效应判断；② 资金与题材主线；③ 明日需关注的方向与风险。"
|
||
"分段清晰。\n\n" + facts)
|
||
text = llm.ask(prompt, temperature=0.6)
|
||
return {"ok": True, "source": "llm", "date": ctx["date"].isoformat(), "facts": facts, "text": text}
|
||
except Exception as e:
|
||
pass
|
||
# 规则降级
|
||
tone = "情绪偏暖、赚钱效应尚可" if (s and s.up > s.down) else "情绪偏弱、需控制仓位"
|
||
text = (f"今日{tone}。{senti_line}。\n"
|
||
f"领涨主线集中在 {', '.join(n for n,_ in ctx['top'][:3])}，"
|
||
f"主力资金净流入居前的是 {', '.join(n for n,_ in ctx['inflow'][:3])}；"
|
||
f"而 {', '.join(n for n,_ in ctx['outflow'][:2])} 遭资金流出，需回避。\n"
|
||
f"明日关注领涨板块的持续性与量能配合，注意高位股退潮风险。\n{DISCLAIMER}")
|
||
return {"ok": True, "source": "rule", "date": ctx["date"].isoformat(), "facts": facts, "text": text}
|
||
|
||
|
||
# ============ 个股诊断 ============
|
||
def _stock_context(symbol):
|
||
with get_session() as s:
|
||
m = s.get(StockMetric, symbol)
|
||
sec = s.get(Security, symbol)
|
||
rows = s.execute(select(DailyQuote).where(DailyQuote.code == symbol)
|
||
.order_by(DailyQuote.date.desc()).limit(60)).scalars().all()
|
||
return m, sec, list(reversed(rows))
|
||
|
||
|
||
def _build_evidence(m, stats, news_tone):
|
||
"""构造证据链：每条含 维度/事实/方向/历史命中率。"""
|
||
def hr(key):
|
||
s = stats.get(key)
|
||
return (s["win_rate"], s["avg_ret"], s["samples"]) if s else (None, None, None)
|
||
|
||
ev = []
|
||
# 趋势
|
||
wr, ar, ns = hr("ma_bull")
|
||
ev.append({"dim": "趋势", "fact": f"均线{'多头排列(MA5>MA10>MA20)' if m.ma_bull else '未呈多头'}，20日{m.ret20:+.1f}%",
|
||
"signal": "bull" if m.ma_bull else ("bear" if m.ret20 < -3 else "neutral"),
|
||
"win_rate": wr, "avg_ret": ar, "samples": ns, "weight": 1.2})
|
||
# 技术(MACD)
|
||
wr, ar, ns = hr("macd_gold")
|
||
ev.append({"dim": "技术", "fact": f"{'MACD金叉' if m.macd_gold else 'MACD未金叉'}，RSI14={m.rsi14:.0f}"
|
||
+ ("（超买）" if m.rsi14 >= 80 else ("（超卖）" if m.rsi14 < 30 else "")),
|
||
"signal": "bull" if m.macd_gold else ("bear" if m.rsi14 >= 80 else "neutral"),
|
||
"win_rate": wr, "avg_ret": ar, "samples": ns, "weight": 1.0})
|
||
# 动量
|
||
wr, ar, ns = hr("up_streak3")
|
||
streak_sig = "bull" if m.up_streak >= 3 else ("bear" if m.ret5 < -5 else "neutral")
|
||
ev.append({"dim": "动量", "fact": f"连涨{m.up_streak}日，5日{m.ret5:+.1f}%",
|
||
"signal": streak_sig, "win_rate": wr if m.up_streak >= 3 else None,
|
||
"avg_ret": ar if m.up_streak >= 3 else None, "samples": ns if m.up_streak >= 3 else None, "weight": 1.0})
|
||
# 资金
|
||
wr, ar, ns = hr("vol_breakout")
|
||
vol_sig = "bull" if (m.vol_ratio > 2 and m.pct > 0) else ("bear" if (m.vol_ratio > 2 and m.pct < 0) else "neutral")
|
||
ev.append({"dim": "资金", "fact": f"量比{m.vol_ratio}，当日{m.pct:+.2f}%，成交额{m.amount}亿",
|
||
"signal": vol_sig, "win_rate": wr if vol_sig != "neutral" else None,
|
||
"avg_ret": ar if vol_sig != "neutral" else None, "samples": ns if vol_sig != "neutral" else None, "weight": 1.1})
|
||
# 位置
|
||
key = "new_high60" if m.pos60 >= 0.99 else ("rsi_oversold" if m.rsi14 < 30 else None)
|
||
wr, ar, ns = hr(key) if key else (None, None, None)
|
||
pos_sig = "bear" if m.pos60 > 0.92 else ("bull" if m.pos60 < 0.2 else "neutral")
|
||
ev.append({"dim": "位置", "fact": f"处于60日 {m.pos60*100:.0f}% 分位" + ("（创新高）" if m.pos60 >= 0.99 else ("（低位）" if m.pos60 < 0.2 else "")),
|
||
"signal": pos_sig, "win_rate": wr, "avg_ret": ar, "samples": ns, "weight": 0.8})
|
||
# 消息面（RAG）
|
||
ev.append({"dim": "消息", "fact": f"近期资讯情绪：{news_tone}",
|
||
"signal": {"利好": "bull", "利空": "bear"}.get(news_tone, "neutral"),
|
||
"win_rate": None, "avg_ret": None, "samples": None, "weight": 0.9})
|
||
return ev
|
||
|
||
|
||
def _confidence_direction(ev):
|
||
sval = {"bull": 1, "bear": -1, "neutral": 0}
|
||
net = sum(sval[e["signal"]] * e["weight"] for e in ev)
|
||
tw = sum(e["weight"] for e in ev if e["signal"] != "neutral")
|
||
agreement = abs(net) / tw if tw else 0.0
|
||
conf = int(max(5, min(96, 35 + agreement * 55 + (8 if tw >= 3 else 0))))
|
||
thr = 0.15 * sum(e["weight"] for e in ev)
|
||
direction = "up" if net > thr else ("down" if net < -thr else "flat")
|
||
return conf, direction, round(net, 2)
|
||
|
||
|
||
def diagnose(symbol):
|
||
m, sec, rows = _stock_context(symbol)
|
||
if not m:
|
||
return {"ok": False, "msg": "该股票无因子数据，请先入库（数据中台→全市场回填或指定入库）"}
|
||
name = (sec.name if sec else m.name) or symbol
|
||
|
||
# 规则打分（0~100）
|
||
trend = 50 + (15 if m.ma_bull else -10) + min(20, max(-20, m.ret20))
|
||
momentum = 50 + min(25, max(-25, m.ret5)) + (10 if 50 < m.rsi14 < 75 else (-10 if m.rsi14 >= 80 else 0))
|
||
capital = 50 + min(30, (m.vol_ratio - 1) * 20) * (1 if m.pct > 0 else -1)
|
||
position = 100 - abs(m.pos60 * 100 - 45) # 价格分位适中得分高
|
||
clamp = lambda x: int(max(5, min(98, x)))
|
||
scores = {"趋势": clamp(trend), "动量": clamp(momentum), "资金": clamp(capital), "位置": clamp(position)}
|
||
total = clamp(sum(scores.values()) / 4)
|
||
|
||
# 证据链 + 历史命中率 + RAG 资讯
|
||
stats = signals.get_stats(horizon=5)
|
||
rctx = rag.stock_context(symbol, limit=5)
|
||
evidence = _build_evidence(m, stats, rctx["tone"])
|
||
confidence, direction, net = _confidence_direction(evidence)
|
||
|
||
# 留痕：本次诊断写入预测，N 日后核验
|
||
try:
|
||
signals.record_prediction(symbol, name, m.date, total, confidence, direction, m.close, horizon=5)
|
||
except Exception:
|
||
pass
|
||
|
||
facts = (
|
||
f"股票：{name}({symbol})\n"
|
||
f"最新价：{m.close}，当日{m.pct:+.2f}%\n"
|
||
f"均线：MA5={m.ma5} MA10={m.ma10} MA20={m.ma20} MA60={m.ma60}，{'多头排列' if m.ma_bull else '非多头'}\n"
|
||
f"涨幅：5日{m.ret5:+.2f}% / 20日{m.ret20:+.2f}% / 60日{m.ret60:+.2f}%\n"
|
||
f"量比：{m.vol_ratio}，RSI14：{m.rsi14}，60日价格分位：{m.pos60*100:.0f}%，"
|
||
f"{'MACD金叉' if m.macd_gold else 'MACD未金叉'}，连涨{m.up_streak}日，成交额{m.amount}亿\n"
|
||
)
|
||
hit_lines = "\n".join(
|
||
f"- {e['dim']}信号「{e['fact']}」历史5日上涨概率 {e['win_rate']}%、平均{e['avg_ret']:+.2f}%（样本{e['samples']}）"
|
||
for e in evidence if e.get("win_rate") is not None)
|
||
base = {"ok": True, "symbol": symbol, "name": name, "scores": scores, "total": total,
|
||
"confidence": confidence, "direction": direction, "evidence": evidence,
|
||
"news": rctx["items"], "news_tone": rctx["tone"], "facts": facts}
|
||
|
||
if llm.enabled():
|
||
try:
|
||
prompt = ("请基于以下个股量化数据、信号历史命中率与相关资讯，输出结构化诊断：\n"
|
||
"1) 一句话结论（含方向与置信判断）；2) 技术面/资金面/趋势面分别点评（引用历史命中率）；"
|
||
"3) 操作建议（含参考性关注位与止损思路）；4) 主要风险。200~320字。\n\n"
|
||
f"{facts}\n历史命中率：\n{hit_lines or '（暂无回测数据，请先在AI准确率页计算）'}\n\n"
|
||
f"{rctx['block'] or '近期无相关资讯。'}\n")
|
||
text = llm.ask(prompt, temperature=0.4)
|
||
base.update({"source": "llm", "text": text})
|
||
return base
|
||
except Exception:
|
||
pass
|
||
|
||
judge = "偏强" if total >= 60 else ("偏弱" if total < 45 else "中性")
|
||
dir_cn = {"up": "看多", "down": "看空", "flat": "中性观望"}[direction]
|
||
text = (f"综合评分 {total}（{judge}），方向{dir_cn}，置信度 {confidence}%。"
|
||
f"趋势{'多头向上' if m.ma_bull else '尚未走强'}，20日{m.ret20:+.1f}%；"
|
||
f"动量 RSI {m.rsi14:.0f}{'，超买防回调' if m.rsi14>=80 else ''}；"
|
||
f"资金{'放量' if m.vol_ratio>=1.5 else '量能一般'}（量比{m.vol_ratio}）；"
|
||
f"价格处于60日 {m.pos60*100:.0f}% 分位；消息面{rctx['tone']}。\n"
|
||
+ (f"依据信号历史表现：\n{hit_lines}\n" if hit_lines else "")
|
||
+ f"操作：{'回踩不破MA20可关注' if m.ma_bull else '等待站稳均线再观察'}，跌破MA20或前低则减仓。"
|
||
f"风险：{'高位放量谨防见顶' if m.pos60>0.9 else '大盘系统性波动'}。\n{DISCLAIMER}")
|
||
base.update({"source": "rule", "text": text})
|
||
return base
|
||
|
||
|
||
# ============ 今日策略 ============
|
||
def today_strategy():
|
||
with get_session() as s:
|
||
d = s.execute(select(func.max(SectorDaily.date))).scalar()
|
||
secs = s.execute(select(SectorDaily).where(SectorDaily.date == d).order_by(SectorDaily.pct.desc()).limit(5)).scalars().all() if d else []
|
||
flows = s.execute(select(FundFlowDaily).where(FundFlowDaily.date == d).order_by(FundFlowDaily.net.desc()).limit(5)).scalars().all() if d else []
|
||
strong = s.execute(select(StockMetric).where(StockMetric.up_streak >= 3).order_by(StockMetric.up_streak.desc()).limit(8)).scalars().all()
|
||
macd = s.execute(select(StockMetric).where(StockMetric.macd_gold.is_(True)).order_by(StockMetric.ret5.desc()).limit(8)).scalars().all()
|
||
facts = (
|
||
f"领涨板块：{', '.join(f'{x.name}({x.pct:+.2f}%)' for x in secs)}\n"
|
||
f"主力净流入板块：{', '.join(f'{x.name}({x.net:+.1f}亿)' for x in flows)}\n"
|
||
f"强势连涨个股：{', '.join(f'{x.name}({x.up_streak}连阳)' for x in strong)}\n"
|
||
f"MACD金叉个股：{', '.join(x.name for x in macd)}\n"
|
||
)
|
||
if llm.enabled():
|
||
try:
|
||
prompt = ("请基于以下盘面数据，给出『今天/接下来怎么做』的策略观点："
|
||
"① 当前资金与题材主线；② 值得关注的方向（给理由）；③ 需要回避的风险。"
|
||
"220字以内。\n\n" + facts)
|
||
text = llm.ask(prompt, temperature=0.6)
|
||
return {"ok": True, "source": "llm", "facts": facts, "text": text}
|
||
except Exception:
|
||
pass
|
||
main = "、".join(x.name for x in secs[:3]) or "暂无数据"
|
||
text = (f"当前资金主线集中在 {main}。"
|
||
f"可重点跟踪净流入居前的 {', '.join(x.name for x in flows[:3])} 板块内的强势个股；"
|
||
f"强势连涨股需注意分歧与退潮节奏，避免追高。\n{DISCLAIMER}")
|
||
return {"ok": True, "source": "rule", "facts": facts, "text": text}
|