fix bug

plot_comparison.py

@@ -220,50 +220,67 @@ def apply_normalization(df, key_cols, normalize_configs, group_cols):
         
         if effective_group_cols:
             def normalize_group(group):
-                base_mask = pd.Series(True, index=group.index)
-                for k, v in base_dict.items():
-                    if k in group.columns:
-                        base_mask = base_mask & (group[k] == v)
-                
-                if base_mask.any():
-                    base_val = group.loc[base_mask, col].mean()
-                    if base_val == 0 or pd.isna(base_val):
-                        valid_data = group[group[col].notna() & (group[col] != 0)]
-                        if len(valid_data) > 0:
-                            base_val = valid_data[col].iloc[0]
-                        else:
-                            return group
-                else:
+                if not base_dict:
                     valid_data = group[group[col].notna() & (group[col] != 0)]
                     if len(valid_data) > 0:
                         base_val = valid_data[col].iloc[0]
                     else:
                         return group
+                else:
+                    base_mask = pd.Series(True, index=group.index)
+                    for k, v in base_dict.items():
+                        if k in group.columns:
+                            base_mask = base_mask & (group[k] == v)
+                    
+                    if base_mask.any():
+                        base_val = group.loc[base_mask, col].mean()
+                        if base_val == 0 or pd.isna(base_val):
+                            valid_data = group[group[col].notna() & (group[col] != 0)]
+                            if len(valid_data) > 0:
+                                base_val = valid_data[col].iloc[0]
+                            else:
+                                return group
+                    else:
+                        valid_data = group[group[col].notna() & (group[col] != 0)]
+                        if len(valid_data) > 0:
+                            base_val = valid_data[col].iloc[0]
+                        else:
+                            return group
                 
                 if base_val == 0 or pd.isna(base_val):
                     return group
                 group[col] = (group[col] / base_val) * 100
                 return group
             df = df.groupby(effective_group_cols, group_keys=False).apply(normalize_group)
-            print(f"  归一化列 '{col}': 每个分组内基准值 {base_dict} = 100%")
+            base_display = base_dict if base_dict else '(自动选择第一个)'
+            print(f"  归一化列 '{col}': 每个分组内基准值 {base_display} = 100%")
         else:
-            base_mask = pd.Series(True, index=df.index)
-            for k, v in base_dict.items():
-                if k in df.columns:
-                    base_mask = base_mask & (df[k] == v)
-            if not base_mask.any():
-                print(f"  警告: 未找到基准值 {base_dict}，跳过归一化")
-                continue
-            base_values = df.loc[base_mask, col].mean()
-            if base_values == 0 or pd.isna(base_values):
+            if not base_dict:
                 valid_data = df[df[col].notna() & (df[col] != 0)]
                 if len(valid_data) > 0:
                     base_values = valid_data[col].iloc[0]
                 else:
                     print(f"  警告: 无有效数据，跳过归一化")
                     continue
+            else:
+                base_mask = pd.Series(True, index=df.index)
+                for k, v in base_dict.items():
+                    if k in df.columns:
+                        base_mask = base_mask & (df[k] == v)
+                if not base_mask.any():
+                    print(f"  警告: 未找到基准值 {base_dict}，跳过归一化")
+                    continue
+                base_values = df.loc[base_mask, col].mean()
+                if base_values == 0 or pd.isna(base_values):
+                    valid_data = df[df[col].notna() & (df[col] != 0)]
+                    if len(valid_data) > 0:
+                        base_values = valid_data[col].iloc[0]
+                    else:
+                        print(f"  警告: 无有效数据，跳过归一化")
+                        continue
             df[col] = (df[col] / base_values) * 100
-            print(f"  归一化列 '{col}': 基准值 {base_dict} = 100%")
+            base_display = base_dict if base_dict else '(自动选择第一个)'
+            print(f"  归一化列 '{col}': 基准值 {base_display} = 100%")
     
     return df
 
@@ -293,17 +310,28 @@ def generate_chart(df_subset, output_path, colkey, outer_group_cols, inner_group
     engine_values = df_grouped[compare_col].unique()
     n_engines = len(engine_values)
     
+    max_inner_groups = 0
+    for _, outer_row in outer_values.iterrows():
+        df_outer = df_grouped.copy()
+        for gcol in outer_group_cols:
+            df_outer = df_outer[df_outer[gcol] == outer_row[gcol]]
+        pt = df_outer.pivot_table(index=inner_group_cols, columns=compare_col, values=metric_cols[0]).fillna(0)
+        max_inner_groups = max(max_inner_groups, len(pt))
+    
     color_palette = ['#2E86AB', '#A23B72', '#F18F01', '#C73E1D', '#3B1F2B', '#95C623', '#7B2D26']
     colors = [color_palette[i % len(color_palette)] for i in range(n_engines)]
     
     normalized_cols = [c['column'] for c in (normalize_configs or [])]
     
     if merge_groups and n_outer > 1:
-        fig, axes = plt.subplots(1, 4, figsize=(8 * n_outer + 20, 10))
+        total_bars = max_inner_groups * n_engines
+        unit_width = 1.0
+        fig_width = max(8 * n_outer + 20, total_bars * unit_width + 5)
+        fig, axes = plt.subplots(1, 4, figsize=(fig_width, 10))
         
-        bar_width = 0.12
-        bar_spacing = 0.05
-        group_gap = 3
+        bar_width = min(0.2, 0.7 / max(n_engines, 1))
+        bar_spacing = bar_width * 0.3
+        group_gap = max(1.5, n_engines * bar_width * 0.8)
         
         x_labels_all = None
         
@@ -381,14 +409,17 @@ def generate_chart(df_subset, output_path, colkey, outer_group_cols, inner_group
                 ax.set_ylabel(metric, fontsize=10)
                 ax.set_title(f'{metric}', fontsize=12, fontweight='bold')
             ax.grid(axis='y', alpha=0.3, linestyle='--')
-            ax.legend(fontsize=5, loc='upper right', framealpha=0.9, ncol=1)
+            ax.legend(fontsize=5, loc='upper left', bbox_to_anchor=(1.02, 1), framealpha=0.9, ncol=1)
     else:
-        fig, axes = plt.subplots(n_outer, 4, figsize=(24, 5 * n_outer))
+        total_bars = max_inner_groups * n_engines
+        unit_width = 0.8
+        fig_width = max(24, total_bars * unit_width + 4)
+        fig, axes = plt.subplots(n_outer, 4, figsize=(fig_width, 5 * n_outer))
         
         if n_outer == 1:
             axes = axes.reshape(1, -1)
         
-        bar_width = 0.2
+        bar_width = min(0.25, 0.8 / max(n_engines, 1))
         
         outer_label = '/'.join(outer_group_cols) if outer_group_cols else '全部'
         
@@ -438,9 +469,9 @@ def generate_chart(df_subset, output_path, colkey, outer_group_cols, inner_group
                 ax.set_xticklabels(x_labels, rotation=45, ha='right', fontsize=7)
                 ax.grid(axis='y', alpha=0.3, linestyle='--')
                 
-                ax.legend(fontsize=6, loc='upper right', framealpha=0.9, ncol=1)
+                ax.legend(fontsize=6, loc='upper left', bbox_to_anchor=(1.02, 1), framealpha=0.9, ncol=1)
     
-    plt.tight_layout()
+    plt.tight_layout(rect=[0, 0, 0.85, 1])
     plt.savefig(output_path, dpi=150, bbox_inches='tight', facecolor='white')
     plt.close()
     return True