Skip to content

Graph-Native Infrastructure for Context and Accountable AI Systems

fix: HybridSearch.search() crashes with AttributeError on non-inmemory backends

KaifAhmad1b4f820568aa7
diff --git a/semantica/vector_store/hybrid_search.py b/semantica/vector_store/hybrid_search.py
index f7788da6..edfb8ccd 100644
--- a/semantica/vector_store/hybrid_search.py
+++ b/semantica/vector_store/hybrid_search.py
@@ -301,6 +301,14 @@ class HybridSearch:
# Handle legacy argument top_k
if "top_k" in options:
k = options["top_k"]
+ # query_vector is derived from `query` above; drop any stray value
+ # passed in **options so it doesn't collide with that derivation.
+ # (**options is a fresh dict per call, so this can't affect the caller.)
+ if "query_vector" in options:
+ self.logger.debug(
+ "Ignoring 'query_vector' passed in options; it is derived from 'query' instead."
+ )
+ options.pop("query_vector")
tracking_id = self.progress_tracker.start_tracking(
module="vector_store",
@@ -309,23 +317,8 @@ class HybridSearch:
)
try:
- # Resolve vector store data if not provided
- if vectors is None and self.vector_store:
- vector_ids = list(self.vector_store.vectors.keys())
- vectors = [self.vector_store.vectors[vid] for vid in vector_ids]
- metadata = [self.vector_store.metadata.get(vid, {}) for vid in vector_ids]
-
- if vectors is None or metadata is None:
- # Check if vectors/metadata are falsy (empty list) but not None
- # If they are None, we can't proceed. If they are empty lists, we return empty results.
- if vectors is None:
- vectors = []
- if metadata is None:
- metadata = []
- if vector_ids is None:
- vector_ids = []
-
- # Handle string query
+ # Resolve the query vector first: it's needed both for local
+ # search below and for the backend-delegated path.
if isinstance(query, str):
self.progress_tracker.update_tracking(
tracking_id, message="Generating query embedding..."
@@ -336,13 +329,78 @@ class HybridSearch:
self.embedding_generator = EmbeddingGenerator()
except (ImportError, OSError):
raise ImportError("EmbeddingGenerator not available for string queries")
-
+
query_vector = self.embedding_generator.generate_embeddings(query, data_type="text")
# Handle if it returns batch (2D) or single (1D)
if len(query_vector.shape) == 2:
query_vector = query_vector[0]
else:
- query_vector = query
+ # Backend stores (e.g. FAISSStore.search_similar) assume an
+ # ndarray and call .ndim on it, so normalize list input here
+ # rather than relying on _vector_search's own conversion,
+ # which the backend-delegated path below bypasses.
+ query_vector = np.asarray(query) if isinstance(query, list) else query
+
+ # Resolve vector store data if not provided
+ if vectors is None and self.vector_store:
+ if hasattr(self.vector_store, "vectors"):
+ # In-memory backend exposes its corpus directly
+ vector_ids = list(self.vector_store.vectors.keys())
+ vectors = [self.vector_store.vectors[vid] for vid in vector_ids]
+ metadata = [self.vector_store.metadata.get(vid, {}) for vid in vector_ids]
+ else:
+ # Backend-agnostic path: other backends (faiss, weaviate,
+ # qdrant, milvus, pinecone, pgvector, sqlite) don't expose
+ # a raw `.vectors` dict, so delegate similarity search to
+ # the store's public API instead.
+ self.progress_tracker.update_tracking(
+ tracking_id, message="Performing vector similarity search..."
+ )
+ fetch_k = k * (4 if metadata_filter else 2)
+ backend_results = self.vector_store.search_vectors(
+ query_vector=query_vector, k=fetch_k, **options
+ )
+
+ if metadata_filter:
+ self.progress_tracker.update_tracking(
+ tracking_id, message="Filtering by metadata..."
+ )
+ backend_results = [
+ result
+ for result in backend_results
+ if metadata_filter.matches(result.get("metadata", {}))
+ ]
+
+ # Normalize to the same result schema the local search
+ # path produces, since backends don't all return the
+ # same set of keys.
+ normalized_results = [
+ {
+ "id": result.get("id"),
+ "score": result.get("score", 0.0),
+ "distance": result.get("distance", result.get("score", 0.0)),
+ "metadata": result.get("metadata", {}),
+ }
+ for result in backend_results[:k]
+ ]
+
+ self.progress_tracker.stop_tracking(
+ tracking_id,
+ status="completed",
+ message=f"Hybrid search completed: {len(normalized_results)} results",
+ )
+ return normalized_results
+
+ if vectors is None:
+ vectors = []
+ if metadata is None:
+ metadata = []
+ if vector_ids is None:
+ # IDs weren't provided alongside explicit vectors/metadata;
+ # generate stable positional ones so downstream indexing
+ # (e.g. filtered_ids[i], _vector_search's vector_ids[idx])
+ # doesn't operate on None.
+ vector_ids = [f"vec_{i}" for i in range(len(vectors))]
# Check if vectors/metadata are empty
# Handle both list and numpy array cases safely
diff --git a/semantica/vector_store/vector_store.py b/semantica/vector_store/vector_store.py
index 1ecd7d36..1348162a 100644
--- a/semantica/vector_store/vector_store.py
+++ b/semantica/vector_store/vector_store.py
@@ -496,8 +496,8 @@ class VectorStore:
# Some stores have store_vectors method
return self._backend_store.store_vectors(vectors, metadata=metadata, **options)
else:
- # Basic add_vectors without metadata
- return self._backend_store.add_vectors(vectors, **options)
+ # add_vectors accepts metadata directly (e.g. FAISSStore)
+ return self._backend_store.add_vectors(vectors, metadata=metadata, **options)
else:
raise NotImplementedError(f"Backend store {type(self._backend_store).__name__} does not have add or add_vectors method")