fix bug evaluation code

app.py

@@ -419,21 +419,33 @@ def search_kbli(text: str):
     html += "</div>"
     return html
 
-def calculate_mrr(retrieved_kodes, relevant_kodes):
+def calculate_mrr(retrieved_kodes, relevant_kodes_set):
     for i, kode in enumerate(retrieved_kodes):
-        if kode in relevant_kodes: return 1.0 / (i + 1)
+        if kode in relevant_kodes_set:
+            return 1.0 / (i + 1)
     return 0.0
 
-def calculate_recall(retrieved_kodes, relevant_kodes, k=10):
+def calculate_recall(retrieved_kodes, relevant_kodes_set, k=10):
     retrieved_k_set = set(retrieved_kodes[:k])
-    relevant_set = set(relevant_kodes)
-    if not relevant_set: return 0.0
-    return len(retrieved_k_set.intersection(relevant_set)) / len(relevant_set)
+
+    if not relevant_kodes_set:
+        return 0.0
+
+    return len(retrieved_k_set & relevant_kodes_set) / len(relevant_kodes_set)
 
 def calculate_ndcg(retrieved_kodes, relevance_dict, k=10):
-    dcg = sum(relevance_dict.get(k, 0) / math.log2(i + 2) for i, k in enumerate(retrieved_kodes[:k]))
+
+    dcg = 0
+    for i, kode in enumerate(retrieved_kodes[:k]):
+        rel = relevance_dict.get(kode, 0)
+        dcg += rel / math.log2(i + 2)
+
     ideal_rels = sorted(relevance_dict.values(), reverse=True)[:k]
-    idcg = sum(rel / math.log2(i + 2) for i, rel in enumerate(ideal_rels))
+
+    idcg = 0
+    for i, rel in enumerate(ideal_rels):
+        idcg += rel / math.log2(i + 2)
+
     return dcg / idcg if idcg > 0 else 0.0
 
 def run_evaluation(file_obj, scenario):
@@ -475,7 +487,7 @@ def run_evaluation(file_obj, scenario):
         retrieved_kodes = [str(r.get('kode')) for r in candidates]
         
         rel_dict = ground_truth.get(q_id, {})
-        relevant_kodes = list(rel_dict.keys())
+        relevant_kodes_set = {k for k, r in rel_dict.items() if r > 0}
         
         mrr = calculate_mrr(retrieved_kodes, relevant_kodes)
         recall = calculate_recall(retrieved_kodes, relevant_kodes, k=10)