Fix MongoDB vector search - remove unsupported relevance_score_fn parameter

main.py

@@ -176,12 +176,11 @@ class MongoDBAtlasVectorStore:
             separators=["\n\n", "\n", " ", ""]
         )
         
-        # Initialize LangChain MongoDB Atlas Vector Search with default metric
+        # Initialize LangChain MongoDB Atlas Vector Search
         self.vector_store = MongoDBAtlasVectorSearch(
             collection=self.collection,
             embedding=self.embeddings,
-            index_name=config.VECTOR_INDEX_NAME,
-            relevance_score_fn=config.DEFAULT_METRIC
+            index_name=config.VECTOR_INDEX_NAME
         )
         
         # Test connection
@@ -235,49 +234,28 @@ class MongoDBAtlasVectorStore:
     def similarity_search(self, query: str, k: int = 3, metric: str = None, score_threshold: float = 0.0) -> List[Tuple[Document, float]]:
         """Perform similarity search using MongoDB Atlas Vector Search with specified metric."""
         try:
-            # Use the provided metric or default
+            # Log the metric being used
             if metric and metric in config.SUPPORTED_METRICS:
-                search_metric = config.SUPPORTED_METRICS[metric]
-                logger.info(f"🔍 Using similarity metric: {metric} -> {search_metric}")
+                logger.info(f"� Using similarity metric: {metric}")
             else:
-                search_metric = config.DEFAULT_METRIC
-                logger.info(f"🔍 Using default similarity metric: {search_metric}")
+                logger.info(f"� Using default similarity metric: cosine")
             
-            # Create a new vector store instance with the specified metric if different from current
-            if search_metric != self.vector_store.relevance_score_fn:
-                temp_vector_store = MongoDBAtlasVectorSearch(
-                    collection=self.collection,
-                    embedding=self.embeddings,
-                    index_name=config.VECTOR_INDEX_NAME,
-                    relevance_score_fn=search_metric
-                )
-                
-                # Use the temporary vector store for this search
-                results = temp_vector_store.similarity_search_with_score(
-                    query=query,
-                    k=k,
-                    # score_threshold=score_threshold
-                )
-            else:
-                # Use the existing vector store
-                results = self.vector_store.similarity_search_with_score(
-                    query=query,
-                    k=k,
-                    # score_threshold=score_threshold
-                )
+            # Perform similarity search using the vector store
+            results = self.vector_store.similarity_search_with_score(
+                query=query,
+                k=k
+            )
             
-            logger.info(f"📊 Found {len(results)} relevant documents for query using {search_metric}")
+            logger.info(f"📊 Found {len(results)} relevant documents for query")
             return results
             
         except Exception as e:
-            logger.error(f"❌ Error performing similarity search with metric {metric}: {e}")
-            # Fallback to default metric if there's an error
+            logger.error(f"❌ Error performing similarity search: {e}")
+            # Fallback to basic similarity search without score
             try:
-                results = self.vector_store.similarity_search_with_score(
-                    query=query,
-                    k=k,
-                    # score_threshold=score_threshold
-                )
+                docs = self.vector_store.similarity_search(query=query, k=k)
+                # Convert to tuple format with dummy scores
+                results = [(doc, 0.0) for doc in docs]
                 logger.info(f"📊 Fallback search returned {len(results)} results")
                 return results
             except Exception as fallback_e: