Skip to content

The agent engineering platform.

retrievers: MultiVectorRetriever similarity_score_threshold search type (#23539)

wenngonga00103731953
diff --git a/libs/langchain/langchain/retrievers/multi_vector.py b/libs/langchain/langchain/retrievers/multi_vector.py
index 2ac989914a..54a4d935dc 100644
--- a/libs/langchain/langchain/retrievers/multi_vector.py
+++ b/libs/langchain/langchain/retrievers/multi_vector.py
@@ -19,6 +19,8 @@ class SearchType(str, Enum):
similarity = "similarity"
"""Similarity search."""
+ similarity_score_threshold = "similarity_score_threshold"
+ """Similarity search with a score threshold."""
mmr = "mmr"
"""Maximal Marginal Relevance reranking of similarity search."""
@@ -64,6 +66,13 @@ class MultiVectorRetriever(BaseRetriever):
sub_docs = self.vectorstore.max_marginal_relevance_search(
query, **self.search_kwargs
)
+ elif self.search_type == SearchType.similarity_score_threshold:
+ sub_docs_and_similarities = (
+ self.vectorstore.similarity_search_with_relevance_scores(
+ query, **self.search_kwargs
+ )
+ )
+ sub_docs = [sub_doc for sub_doc, _ in sub_docs_and_similarities]
else:
sub_docs = self.vectorstore.similarity_search(query, **self.search_kwargs)
@@ -89,6 +98,13 @@ class MultiVectorRetriever(BaseRetriever):
sub_docs = await self.vectorstore.amax_marginal_relevance_search(
query, **self.search_kwargs
)
+ elif self.search_type == SearchType.similarity_score_threshold:
+ sub_docs_and_similarities = (
+ await self.vectorstore.asimilarity_search_with_relevance_scores(
+ query, **self.search_kwargs
+ )
+ )
+ sub_docs = [sub_doc for sub_doc, _ in sub_docs_and_similarities]
else:
sub_docs = await self.vectorstore.asimilarity_search(
query, **self.search_kwargs
diff --git a/libs/langchain/tests/unit_tests/retrievers/test_multi_vector.py b/libs/langchain/tests/unit_tests/retrievers/test_multi_vector.py
index e35244c77d..2fdc8009fb 100644
--- a/libs/langchain/tests/unit_tests/retrievers/test_multi_vector.py
+++ b/libs/langchain/tests/unit_tests/retrievers/test_multi_vector.py
@@ -1,13 +1,20 @@
-from typing import Any, List
+from typing import Any, Callable, List, Tuple
from langchain_core.documents import Document
-from langchain.retrievers.multi_vector import MultiVectorRetriever
+from langchain.retrievers.multi_vector import MultiVectorRetriever, SearchType
from langchain.storage import InMemoryStore
from tests.unit_tests.indexes.test_indexing import InMemoryVectorStore
class InMemoryVectorstoreWithSearch(InMemoryVectorStore):
+ @staticmethod
+ def _identity_fn(score: float) -> float:
+ return score
+
+ def _select_relevance_score_fn(self) -> Callable[[float], float]:
+ return self._identity_fn
+
def similarity_search(
self, query: str, k: int = 4, **kwargs: Any
) -> List[Document]:
@@ -16,6 +23,14 @@ class InMemoryVectorstoreWithSearch(InMemoryVectorStore):
return []
return [res]
+ def similarity_search_with_score(
+ self, query: str, k: int = 4, **kwargs: Any
+ ) -> List[Tuple[Document, float]]:
+ res = self.store.get(query)
+ if res is None:
+ return []
+ return [(res, 0.8)]
+
def test_multi_vector_retriever_initialization() -> None:
vectorstore = InMemoryVectorstoreWithSearch()
@@ -41,3 +56,65 @@ async def test_multi_vector_retriever_initialization_async() -> None:
results = await retriever.ainvoke("1")
assert len(results) > 0
assert results[0].page_content == "test document"
+
+
+def test_multi_vector_retriever_similarity_search_with_score() -> None:
+ documents = [Document(page_content="test document", metadata={"doc_id": "1"})]
+ vectorstore = InMemoryVectorstoreWithSearch()
+ vectorstore.add_documents(documents, ids=["1"])
+
+ # score_threshold = 0.5
+ retriever = MultiVectorRetriever( # type: ignore[call-arg]
+ vectorstore=vectorstore,
+ docstore=InMemoryStore(),
+ doc_id="doc_id",
+ search_kwargs={"score_threshold": 0.5},
+ search_type=SearchType.similarity_score_threshold,
+ )
+ retriever.docstore.mset(list(zip(["1"], documents)))
+ results = retriever.invoke("1")
+ assert len(results) == 1
+ assert results[0].page_content == "test document"
+
+ # score_threshold = 0.9
+ retriever = MultiVectorRetriever( # type: ignore[call-arg]
+ vectorstore=vectorstore,
+ docstore=InMemoryStore(),
+ doc_id="doc_id",
+ search_kwargs={"score_threshold": 0.9},
+ search_type=SearchType.similarity_score_threshold,
+ )
+ retriever.docstore.mset(list(zip(["1"], documents)))
+ results = retriever.invoke("1")
+ assert len(results) == 0
+
+
+async def test_multi_vector_retriever_similarity_search_with_score_async() -> None:
+ documents = [Document(page_content="test document", metadata={"doc_id": "1"})]
+ vectorstore = InMemoryVectorstoreWithSearch()
+ await vectorstore.aadd_documents(documents, ids=["1"])
+
+ # score_threshold = 0.5
+ retriever = MultiVectorRetriever( # type: ignore[call-arg]
+ vectorstore=vectorstore,
+ docstore=InMemoryStore(),
+ doc_id="doc_id",
+ search_kwargs={"score_threshold": 0.5},
+ search_type=SearchType.similarity_score_threshold,
+ )
+ await retriever.docstore.amset(list(zip(["1"], documents)))
+ results = retriever.invoke("1")
+ assert len(results) == 1
+ assert results[0].page_content == "test document"
+
+ # score_threshold = 0.9
+ retriever = MultiVectorRetriever( # type: ignore[call-arg]
+ vectorstore=vectorstore,
+ docstore=InMemoryStore(),
+ doc_id="doc_id",
+ search_kwargs={"score_threshold": 0.9},
+ search_type=SearchType.similarity_score_threshold,
+ )
+ await retriever.docstore.amset(list(zip(["1"], documents)))
+ results = retriever.invoke("1")
+ assert len(results) == 0