模型，评分，修复网格策略市场状态监听

2026-06-22 10:48:40 +08:00
parent 2e3202968d
commit f938c453e1
42 changed files with 3803 additions and 108 deletions
@@ -0,0 +1,97 @@
+"""
+大盘独立性特征 (3维) — calculate_independence_features(df)
+使用 HS300(000300) 作为 benchmark，最近 20 个交易日 OLS 回归。
+"""
+import numpy as np
+import pandas as pd
+from core.scoring.config import WINDOW_20D
+from core.scoring.features.v3_2_features import _ols_slope
+
+
+def calculate_independence_features(ctx) -> pd.DataFrame:
+    """计算大盘独立性特征 (3维)"""
+    kline = ctx.kline.copy()
+    hs300 = ctx.hs300_kline.copy() if ctx.hs300_kline is not None else pd.DataFrame()
+
+    if kline.empty or hs300.empty:
+        return pd.DataFrame()
+
+    # 准备 HS300 收益率序列
+    hs300 = hs300.sort_values('trade_date')
+    hs300['market_return'] = hs300['close'].pct_change()
+    hs300['market_amplitude'] = (hs300['high'] - hs300['low']) / hs300['open']
+    hs300 = hs300.dropna(subset=['market_return', 'market_amplitude'])
+
+    # 对齐日期
+    hs300_dates = set(hs300['trade_date'].dt.date
+                      if hasattr(hs300['trade_date'], 'dt') else hs300['trade_date'])
+
+    kline = kline.sort_values(['stock_code', 'trade_date'])
+    kline['trade_date_dt'] = (kline['trade_date'].dt.date
+                              if hasattr(kline['trade_date'], 'dt')
+                              else pd.to_datetime(kline['trade_date']).dt.date)
+
+    candidates = ctx.candidates
+
+    # 计算 HS300 最近20日平均振幅
+    hs300_tail = hs300.tail(WINDOW_20D)
+    hs300_avg_amp = hs300_tail['market_amplitude'].mean() if len(hs300_tail) > 0 else 0
+
+    features = {}
+    grouped = kline.groupby('stock_code')
+
+    for code, group in grouped:
+        if code not in candidates:
+            continue
+        if len(group) < 20:
+            continue
+
+        g = group.sort_values('trade_date').tail(120)
+        closes = g['close'].values
+
+        # 计算个股日收益率
+        stock_rets = np.diff(closes) / np.where(closes[:-1] > 0, closes[:-1], 1)
+
+        # 对齐 HS300 收益率 (取对应日期)
+        # 简化: 取最近 N 个交易日的数据点
+        n = min(WINDOW_20D, len(stock_rets))
+
+        # 获取 HS300 最近 n 天的 market_return
+        market_rets = hs300['market_return'].tail(n + 1).values
+        if len(market_rets) < n:
+            market_rets = hs300['market_return'].values[-n - 1:]
+
+        # 对齐长度
+        min_len = min(n, len(market_rets) - 1, len(stock_rets))
+        if min_len < 5:  # 至少需要5个数据点做回归
+            continue
+
+        stock_ret_window = stock_rets[-min_len:]
+        market_ret_window = market_rets[-min_len:]
+
+        feat = {}
+
+        # OLS 回归: stock_ret ~ market_return
+        try:
+            slope, r_value = _ols_slope(market_ret_window, stock_ret_window)
+            residuals = stock_ret_window - slope * market_ret_window
+
+            # 58. market_residual_volatility_20d: std(residuals) × √252
+            feat['market_residual_volatility_20d'] = np.std(residuals, ddof=1) * np.sqrt(252)
+
+            # 59. market_independence_ratio_20d: 1 - R²
+            feat['market_independence_ratio_20d'] = 1 - r_value ** 2
+
+        except Exception:
+            feat['market_residual_volatility_20d'] = 0
+            feat['market_independence_ratio_20d'] = 1
+
+        # 60. market_amplitude_deviation_20d: 个股平均振幅 - HS300 平均振幅
+        g_amps = (g['high'].values[-20:] - g['low'].values[-20:]) / np.where(
+            g['open'].values[-20:] > 0, g['open'].values[-20:], 1
+        )
+        feat['market_amplitude_deviation_20d'] = np.mean(g_amps) - hs300_avg_amp
+
+        features[code] = feat
+
+    return pd.DataFrame.from_dict(features, orient='index')