Update grace_eval.py

grace_eval.py

@@ -1,8 +1,7 @@
-from evaluate.visualization import radar_plot
 import matplotlib.pyplot as plt
 
 def compute_sample_scores(results, prompt):
-    # 示例人工评分（实际可扩展为平均多个 prompt 的客观指标）
+    # 示例评分 - 实际应用中可以用更复杂的评估逻辑
     return {
         "sd_v1_5": {"G": 4, "R": 4, "A": 4, "E": 3},
         "openjourney_v4": {"G": 3, "R": 4, "A": 5, "E": 3},
@@ -10,6 +9,22 @@ def compute_sample_scores(results, prompt):
     }
 
 def plot_radar(scores_dict, out_path="radar.png"):
-    radar_plot(scores_dict, model_names=list(scores_dict.keys()), title="GRACE Radar Comparison")
-    plt.savefig(out_path)
-    plt.close()
+    categories = list(scores_dict.values())[0].keys()
+    N = len(categories)
+    
+    fig, ax = plt.subplots(figsize=(6, 6), subplot_kw=dict(polar=True))
+    
+    for model, scores in scores_dict.items():
+        values = list(scores.values())
+        values += values[:1]  # 闭合雷达图
+        angles = [n / float(N) * 2 * 3.14159 for n in range(N)]
+        angles += angles[:1]
+        ax.plot(angles, values, linewidth=1, linestyle='solid', label=model)
+        ax.fill(angles, values, alpha=0.1)
+    
+    ax.set_xticks(angles[:-1])
+    ax.set_xticklabels(categories)
+    ax.set_title("GRACE 评估雷达图 (CPU模式)", size=12, y=1.1)
+    ax.legend(loc='upper right')
+    plt.savefig(out_path, bbox_inches='tight')
+    plt.close()