Skip to content

The agent engineering platform.

qdrant: Add `similarity_search_with_score_by_vector()` function to the `QdrantVectorStore` (#29641)

Vincent Emonet3645181d0ea3
diff --git a/libs/partners/qdrant/langchain_qdrant/qdrant.py b/libs/partners/qdrant/langchain_qdrant/qdrant.py
index b21dae2efe..964bf209f9 100644
--- a/libs/partners/qdrant/langchain_qdrant/qdrant.py
+++ b/libs/partners/qdrant/langchain_qdrant/qdrant.py
@@ -568,7 +568,7 @@ class QdrantVectorStore(VectorStore):
for result in results
]
- def similarity_search_by_vector(
+ def similarity_search_with_score_by_vector(
self,
embedding: List[float],
k: int = 4,
@@ -578,11 +578,11 @@ class QdrantVectorStore(VectorStore):
score_threshold: Optional[float] = None,
consistency: Optional[models.ReadConsistency] = None,
**kwargs: Any,
- ) -> List[Document]:
+ ) -> List[tuple[Document, float]]:
"""Return docs most similar to embedding vector.
Returns:
- List of Documents most similar to the query.
+ List of Documents most similar to the query and distance for each.
"""
qdrant_filter = filter
@@ -609,15 +609,46 @@ class QdrantVectorStore(VectorStore):
).points
return [
- self._document_from_point(
- result,
- self.collection_name,
- self.content_payload_key,
- self.metadata_payload_key,
+ (
+ self._document_from_point(
+ result,
+ self.collection_name,
+ self.content_payload_key,
+ self.metadata_payload_key,
+ ),
+ result.score,
)
for result in results
]
+ def similarity_search_by_vector(
+ self,
+ embedding: List[float],
+ k: int = 4,
+ filter: Optional[models.Filter] = None,
+ search_params: Optional[models.SearchParams] = None,
+ offset: int = 0,
+ score_threshold: Optional[float] = None,
+ consistency: Optional[models.ReadConsistency] = None,
+ **kwargs: Any,
+ ) -> List[Document]:
+ """Return docs most similar to embedding vector.
+
+ Returns:
+ List of Documents most similar to the query.
+ """
+ results = self.similarity_search_with_score_by_vector(
+ embedding,
+ k,
+ filter=filter,
+ search_params=search_params,
+ offset=offset,
+ score_threshold=score_threshold,
+ consistency=consistency,
+ **kwargs,
+ )
+ return list(map(itemgetter(0), results))
+
def max_marginal_relevance_search(
self,
query: str,
diff --git a/libs/partners/qdrant/tests/integration_tests/qdrant_vector_store/test_search.py b/libs/partners/qdrant/tests/integration_tests/qdrant_vector_store/test_search.py
index 9525c5c17d..162c819afd 100644
--- a/libs/partners/qdrant/tests/integration_tests/qdrant_vector_store/test_search.py
+++ b/libs/partners/qdrant/tests/integration_tests/qdrant_vector_store/test_search.py
@@ -66,6 +66,39 @@ def test_similarity_search_by_vector(
assert_documents_equals(output, [Document(page_content="foo")])
[email protected]("location", qdrant_locations())
[email protected]("content_payload_key", [QdrantVectorStore.CONTENT_KEY, "foo"])
+ "metadata_payload_key", [QdrantVectorStore.METADATA_KEY, "bar"]
+)
[email protected]("vector_name", ["", "my-vector"])
[email protected]("batch_size", [1, 64])
+def test_similarity_search_with_score_by_vector(
+ location: str,
+ content_payload_key: str,
+ metadata_payload_key: str,
+ vector_name: str,
+ batch_size: int,
+) -> None:
+ """Test end to end construction and search."""
+ texts = ["foo", "bar", "baz"]
+ docsearch = QdrantVectorStore.from_texts(
+ texts,
+ ConsistentFakeEmbeddings(),
+ location=location,
+ content_payload_key=content_payload_key,
+ metadata_payload_key=metadata_payload_key,
+ batch_size=batch_size,
+ vector_name=vector_name,
+ )
+ embeddings = ConsistentFakeEmbeddings().embed_query("foo")
+ output = docsearch.similarity_search_with_score_by_vector(embeddings, k=1)
+ assert len(output) == 1
+ document, score = output[0]
+ assert_documents_equals([document], [Document(page_content="foo")])
+ assert score >= 0
+
+
@pytest.mark.parametrize("location", qdrant_locations())
@pytest.mark.parametrize(
"metadata_payload_key", [QdrantVectorStore.METADATA_KEY, "bar"]