From 0b08e5f6cce2accba740f051e2b2fa8c16ebeb74 Mon Sep 17 00:00:00 2001 From: xxm Date: Sat, 15 Aug 2026 20:53:51 +0800 Subject: [PATCH] =?UTF-8?q?perf:=20=E6=B8=A9=E5=8C=BA=E6=95=B0=E6=8D=AE?= =?UTF-8?q?=E6=8C=89=E5=91=A8=E6=9C=9F=E8=AE=A1=E7=AE=97=E5=8A=A0=E7=BC=93?= =?UTF-8?q?=E5=AD=98(=E9=94=AE=3Dperiod=5Ftag+=E6=95=B0=E6=8D=AE=E7=89=88?= =?UTF-8?q?=E6=9C=AC)=E2=80=94=E2=80=94=E7=8E=B0=E5=9C=BA=E5=BD=92?= =?UTF-8?q?=E5=9B=A030=E7=A7=92/=E8=AF=B7=E6=B1=82,=E7=BC=93=E5=AD=98?= =?UTF-8?q?=E5=90=8E=E6=AF=AB=E7=A7=92=E7=BA=A7?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- server.py | 253 +++++++++++++++++++++++++++++------------------------- 1 file changed, 138 insertions(+), 115 deletions(-) diff --git a/server.py b/server.py index 72b8f9db..920f6f13 100644 --- a/server.py +++ b/server.py @@ -39,6 +39,142 @@ def _chk_tcp(host, port, timeout=3): except Exception: return False +# ── 2026-08-15 温区数据按周期计算 + 缓存（现场温区归因约30秒/请求，缓存后毫秒级）── +# 键 = period_tag + strategy_research 最新 created_at（数据更新即失效） +_regime_winrates_cache = {"key": None, "pt": None, "data": None} + + +def _compute_regime_winrates_cached(pt, _approx_univ): + """按 period_tag 从 strategy_research 对应周期 trades 现场温区归因（带缓存）""" + import sqlite3 as _sq + import math as _mth + from collections import defaultdict as _dd + from flask import request as _req + + # 缓存键 + _cache_key = None + try: + _cc = _sq.connect(str(DATA_DIR / "mofin.db"), timeout=5) + _ver_s = _cc.execute("SELECT MAX(created_at) FROM strategy_research").fetchone()[0] + _cc.close() + _cache_key = (pt or "2y") + "|" + str(_ver_s) + except Exception: + _cache_key = None + if _cache_key and _regime_winrates_cache.get("key") == _cache_key \ + and _regime_winrates_cache.get("pt") == (pt or "2y"): + return _regime_winrates_cache.get("data") or {} + + _pt_chain = {'1m': '1y', '6m': '1y', '1y': '1y', '2y': '2y', '5y': '5y', '10y': '10y'} + _pt_use = _pt_chain.get(pt or '2y', '2y') + _regime_winrates = {} + try: + _c = _sq.connect(str(DATA_DIR / "mofin.db"), timeout=10) + _c.execute("PRAGMA busy_timeout=10000") + _row_map = {} + for _r in _c.execute( + "SELECT version, market, period_tag, results_json FROM strategy_research " + "ORDER BY version, market, " + "CASE period_tag WHEN '1y' THEN 1 WHEN '2y' THEN 2 WHEN '5y' THEN 3 WHEN '10y' THEN 4 ELSE 5 END" + ).fetchall(): + _k = (_r[0], _r[1]) + if _k not in _row_map: + _row_map[_k] = {} + _row_map[_k][_r[2]] = _r[3] + _rmap_a = dict(_c.execute("SELECT date, regime FROM market_regime WHERE market='a'").fetchall()) + _rmap_hk = dict(_c.execute("SELECT date, regime FROM market_regime WHERE market='hk'").fetchall()) + _c.close() + + def _calc_extra(_trades): + if not _trades: + return {} + _profits = [t.get("profit_pct", 0) for t in _trades] + _wins = [p for p in _profits if p > 0] + _losses = [p for p in _profits if p <= 0] + _wr = len(_wins) / len(_profits) * 100 if _profits else 0 + _avg = sum(_profits) / len(_profits) if _profits else 0 + _avg_w = sum(_wins) / len(_wins) if _wins else 0 + _avg_l = abs(sum(_losses) / len(_losses)) if _losses else 1 + _pf = _avg_w / _avg_l if _avg_l > 0 else 0 + _mr = _avg / 100 + _std = _mth.sqrt(sum((p / 100 - _mr) ** 2 for p in _profits) / (len(_profits) - 1)) if len(_profits) > 1 else 0 + _sh = _mr / _std * _mth.sqrt(252) if _std > 0 else 0 + _holds = [t.get("hold_days", 0) for t in _trades if t.get("hold_days")] + _ah = sum(_holds) / len(_holds) if _holds else 0 + return {"win_rate": round(_wr, 1), "avg_pnl": round(_avg, 2), + "avg_hold_days": round(_ah, 1), "sharpe_ratio": round(_sh, 2), + "profit_factor": round(_pf, 2)} + + def _portfolio_sim(_trades, _cap=1000000, _slots=10): + if not _trades: + return {} + try: + from strategy_lab import portfolio_sim + return portfolio_sim(_trades, capital=_cap, max_positions=_slots, cost=True) + except Exception: + return {} + + for (_ver, _mkt), _periods in _row_map.items(): + _js = None + _cand = None + for _p in [_pt_use, '5y', '10y']: + if _p in _periods and _periods[_p]: + _cand = _periods[_p] + break + if not _cand: + continue + try: + _trades = json.loads(_cand).get("trades", []) + except Exception: + continue + if not _trades: + continue + _rmap = _rmap_a if _mkt != 'hk' else _rmap_hk + _by_regime = _dd(list) + for _t in _trades: + _ed = _t.get("entry_date", "") + if _ed in _rmap: + _by_regime[_rmap[_ed]].append(_t) + for _reg, _reg_trades in _by_regime.items(): + if len(_reg_trades) < 2: + continue + _extra = _calc_extra(_reg_trades) + _sim = _portfolio_sim(_reg_trades) + if not _sim: + continue + _wr_v = _extra.get("win_rate") + _cagr_v = _sim.get("cagr_pct") + _ret_v = _sim.get("total_return_pct") + _dd_v = _sim.get("portfolio_max_dd_pct") + _cf_v = _sim.get("capital_final") + _pt_v = _sim.get("positions_taken") + _sh_v = _extra.get("sharpe_ratio") + _pf_v = _extra.get("profit_factor") + _regime_winrates.setdefault(_ver, {})[_reg] = { + "trades": len(_reg_trades), + "win_rate": _wr_v, "avg_pnl": _extra.get("avg_pnl"), + "avg_hold_days": _extra.get("avg_hold_days"), + "total_return_pct": _ret_v, "cagr_pct": _cagr_v, + "max_dd_pct": _dd_v, "capital_final": _cf_v, + "positions_taken": _pt_v, "sharpe_ratio": _sh_v, + "profit_factor": _pf_v, + "period_tag": _pt_use, + "portfolio": {"cagr_pct": _cagr_v, "total_return_pct": _ret_v, + "portfolio_max_dd_pct": _dd_v, "capital_final": _cf_v, + "positions_taken": _pt_v, "sharpe_ratio": _sh_v, + "profit_factor": _pf_v}, + "universality": _approx_univ(_ver, _reg, len(_reg_trades)), + } + except Exception: + pass + if _cache_key: + try: + _regime_winrates_cache["key"] = _cache_key + _regime_winrates_cache["pt"] = pt or "2y" + _regime_winrates_cache["data"] = _regime_winrates + except Exception: + pass + return _regime_winrates + def _chk_http(host, port, path, timeout=3): try: @@ -565,122 +701,9 @@ def api_research_strategies(): _active_set = set(_weights.get("active") or []) _hk_market = (_weights.get("markets") or {}).get("hk") or {} _active_set |= set(_hk_market.get("active") or []) - # ── 2026-08-15 温区数据跟随所选周期（period_tag）── - # 病状：旧版读 strategy_regime_perf 全量表（10y/5y 最长窗口），与前端周期下拉无关， - # 导致选"近2年"表内仍是全量数据（老莫抓包发现，严重误导）。 - # 修复：按 period_tag 从 strategy_research 对应周期 trades 现场温区归因（calc_extra+portfolio_sim）， - # 周期记录缺失时回退到最近更长周期（1m/6m→1y，2y→2y，5y→5y，10y→10y）。 - _regime_winrates = {} - try: - import sqlite3 as _sq - import math as _mth - from collections import defaultdict as _dd - _c = _sq.connect(str(DATA_DIR / "mofin.db"), timeout=10) - _c.execute("PRAGMA busy_timeout=10000") - # 周期回退链：前端可选 1m/6m/1y/2y/5y/10y；strategy_research 有 1y/2y/5y/10y 记录 - _pt_chain = {'1m': '1y', '6m': '1y', '1y': '1y', '2y': '2y', '5y': '5y', '10y': '10y'} - _pt_use = _pt_chain.get(pt or '2y', '2y') - # 各策略按周期取 trades（含市场维度；港股策略 market='hk'） - _row_map = {} - for _r in _c.execute( - "SELECT version, market, period_tag, results_json FROM strategy_research " - "ORDER BY version, market, " - "CASE period_tag WHEN '1y' THEN 1 WHEN '2y' THEN 2 WHEN '5y' THEN 3 WHEN '10y' THEN 4 ELSE 5 END" - ).fetchall(): - _k = (_r[0], _r[1]) - if _k not in _row_map: - _row_map[_k] = {} - _row_map[_k][_r[2]] = _r[3] - # 温区映射（A股+港股分开） - _rmap_a = dict(_c.execute("SELECT date, regime FROM market_regime WHERE market='a'").fetchall()) - _rmap_hk = dict(_c.execute("SELECT date, regime FROM market_regime WHERE market='hk'").fetchall()) - _c.close() + # ── 2026-08-15 温区数据按所选周期计算（函数内缓存，替代全量表/内联重算）── + _regime_winrates = _compute_regime_winrates_cached(pt, _approx_regime_universality) - def _calc_extra(_trades): - if not _trades: - return {} - _profits = [t.get("profit_pct", 0) for t in _trades] - _wins = [p for p in _profits if p > 0] - _losses = [p for p in _profits if p <= 0] - _wr = len(_wins) / len(_profits) * 100 if _profits else 0 - _avg = sum(_profits) / len(_profits) if _profits else 0 - _avg_w = sum(_wins) / len(_wins) if _wins else 0 - _avg_l = abs(sum(_losses) / len(_losses)) if _losses else 1 - _pf = _avg_w / _avg_l if _avg_l > 0 else 0 - _mr = _avg / 100 - _std = _mth.sqrt(sum((p / 100 - _mr) ** 2 for p in _profits) / (len(_profits) - 1)) if len(_profits) > 1 else 0 - _sh = _mr / _std * _mth.sqrt(252) if _std > 0 else 0 - _holds = [t.get("hold_days", 0) for t in _trades if t.get("hold_days")] - _ah = sum(_holds) / len(_holds) if _holds else 0 - return {"win_rate": round(_wr, 1), "avg_pnl": round(_avg, 2), - "avg_hold_days": round(_ah, 1), "sharpe_ratio": round(_sh, 2), - "profit_factor": round(_pf, 2)} - - def _portfolio_sim(_trades, _cap=1000000, _slots=10): - if not _trades: - return {} - try: - from strategy_lab import portfolio_sim - return portfolio_sim(_trades, capital=_cap, max_positions=_slots, cost=True) - except Exception: - return {} - - for (_ver, _mkt), _periods in _row_map.items(): - # 取目标周期 trades（缺失则用最近更长周期） - _js = None - _cand = None - for _p in [_pt_use, '5y', '10y']: - if _p in _periods and _periods[_p]: - _cand = _periods[_p] - break - if not _cand: - continue - try: - _trades = json.loads(_cand).get("trades", []) - except Exception: - continue - if not _trades: - continue - _rmap = _rmap_a if _mkt != 'hk' else _rmap_hk - _by_regime = _dd(list) - for _t in _trades: - _ed = _t.get("entry_date", "") - if _ed in _rmap: - _by_regime[_rmap[_ed]].append(_t) - for _reg, _reg_trades in _by_regime.items(): - if len(_reg_trades) < 2: - continue - _extra = _calc_extra(_reg_trades) - _sim = _portfolio_sim(_reg_trades) - if not _sim: - continue - _wr_v = _extra.get("win_rate") - _cagr_v = _sim.get("cagr_pct") - _ret_v = _sim.get("total_return_pct") - _dd_v = _sim.get("portfolio_max_dd_pct") - _cf_v = _sim.get("capital_final") - _pt_v = _sim.get("positions_taken") - _sh_v = _extra.get("sharpe_ratio") - _pf_v = _extra.get("profit_factor") - _regime_winrates.setdefault(_ver, {})[_reg] = { - "trades": len(_reg_trades), - "win_rate": _wr_v, "avg_pnl": _extra.get("avg_pnl"), - "avg_hold_days": _extra.get("avg_hold_days"), - "total_return_pct": _ret_v, "cagr_pct": _cagr_v, - "max_dd_pct": _dd_v, "capital_final": _cf_v, - "positions_taken": _pt_v, "sharpe_ratio": _sh_v, - "profit_factor": _pf_v, - "period_tag": _pt_use, - # 温区级组合级指标（温区行也要显示组合级列） - "portfolio": {"cagr_pct": _cagr_v, "total_return_pct": _ret_v, - "portfolio_max_dd_pct": _dd_v, "capital_final": _cf_v, - "positions_taken": _pt_v, "sharpe_ratio": _sh_v, - "profit_factor": _pf_v}, - # 温区级 universality（近似） - "universality": _approx_regime_universality(_ver, _reg, len(_reg_trades)), - } - except Exception: - pass # 温区级普适前先建立日期→温区映射（A股 market_regime） _regime_map = {} try: