1、新增dp策略

src/analysis/analysis_utils.py

@@ -80,6 +80,15 @@ def calculate_metrics(
     else:
         calmar_ratio = float("inf")
 
+    target_return = 0
+    downside_returns = df_returns[df_returns < target_return]
+    downside_deviation = downside_returns.std()
+
+    sortino_ratio = 0.0
+    if downside_deviation > 0:
+        # 年化超额收益率 / 年化下行波动率
+        sortino_ratio = np.sqrt(252) * (excess_daily_returns.mean() / downside_deviation)
+
     total_commissions = sum(t.commission for t in trades)
 
     # --- 重新计算交易相关指标 ---
@@ -139,6 +148,7 @@ def calculate_metrics(
         "max_drawdown": max_drawdown,
         "sharpe_ratio": sharpe_ratio,
         "calmar_ratio": calmar_ratio,
+        "sortino_ratio": sortino_ratio,
         "total_trades": len(trades),  # All buy and sell trades
         "transaction_costs": total_commissions,
         "total_realized_pnl": total_realized_pnl,  # New

src/analysis/grid_search_analyzer.py

@@ -46,7 +46,7 @@ class GridSearchAnalyzer:
         print(f"  {self.optimization_metric}: {best_result[self.optimization_metric]:.4f}")
         return best_result
 
-    def plot_heatmap(self, title: str = "heatmap"):
+    def plot_heatmap(self, title: str = "heatmap", order_direction: str=None):
         """
         绘制两个参数的热力图。
         """
@@ -59,11 +59,15 @@ class GridSearchAnalyzer:
 
         heatmap_matrix = np.zeros((len(y_values), len(x_values)))
 
+        max_values = max([item[self.optimization_metric] for item in self.grid_results])
         # 填充网格
         for item in self.grid_results:
             x_idx = x_values.index(item[self.param1_name])
             y_idx = y_values.index(item[self.param2_name])
-            heatmap_matrix[y_idx, x_idx] = item[self.optimization_metric]
+            if max_values < 0.1:
+                heatmap_matrix[y_idx, x_idx] = item[self.optimization_metric] * 100
+            else:
+                heatmap_matrix[y_idx, x_idx] = item[self.optimization_metric]
 
         print([x_values[0], x_values[-1], y_values[0], y_values[-1]])
         fig, ax = plt.subplots(figsize=(10, 8))
@@ -76,7 +80,7 @@ class GridSearchAnalyzer:
 
         ax.set_xlabel(self.param1_name)
         ax.set_ylabel(self.param2_name)
-        ax.set_title(f"{title}: {self.optimization_metric} vs. {self.param1_name} & {self.param2_name}")
+        ax.set_title(f"{title}: {self.optimization_metric} vs. {self.param1_name} & {self.param2_name}, order={order_direction}")
 
         cbar = fig.colorbar(im, ax=ax)
         cbar.set_label(self.optimization_metric)

src/analysis/result_analyzer.py

@@ -166,6 +166,23 @@ class ResultAnalyzer:
             # DataFrame 的结构为：['indicator_value', 'realized_pnl']
             df = pd.DataFrame({"indicator_value": indi_values, "realized_pnl": pnls})
 
+            def remove_extreme(df, col='indicator_value', k=3):
+                """IQR 稳健过滤，返回过滤后的 df 和被剔除的行数"""
+                q1, q3 = df[col].quantile([0.25, 0.75])
+                iqr = q3 - q1
+                lower, upper = q1 - k * iqr, q3 + k * iqr
+                mask = df[col].between(lower, upper)
+                n_remove = (~mask).sum()
+                return df[mask].copy(), n_remove
+
+            df, n_drop = remove_extreme(df)  # 默认 k=2.5
+            if n_drop:
+                print(f"指标 '{indicator_name}' 过滤掉 {n_drop} 个极端异常值 "
+                      f"({n_drop / len(df) * 100:.1f}%)，剩余 {len(df)} 条样本。")
+            if df.empty:
+                print(f"指标 '{indicator_name}' 过滤后无数据，跳过绘图。")
+                continue
+
             # 确保数据框不为空
             if df.empty:
                 print(f"指标 '{indicator_name}' 的数据框为空，跳过绘图。")