Archives/langchain
2
0
Fork 1
mirror of https://github.com/hwchase17/langchain synced 2024-11-06 03:20:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revert "add filter to sklearn vector store functions (#8113)" (#8760)

Browse Source
This commit is contained in:
Bagatur 2023-08-04 08:13:32 -07:00 committed by GitHub
parent 6b007e2829
commit 0d5a90f30a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 26 additions and 131 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
docs/extras/integrations/vectorstores/sklearn.ipynb
View File

@ -13,7 +13,7 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
@ -56,7 +56,7 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
@ -65,7 +65,7 @@
"from langchain.vectorstores import SKLearnVectorStore\n",
"from langchain.document_loaders import TextLoader\n",
"\n",
"loader = TextLoader(\"../../../extras/modules/state_of_the_union.txt\")\n",
"loader = TextLoader(\"../../../state_of_the_union.txt\")\n",
"documents = loader.load()\n",
"text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)\n",
"docs = text_splitter.split_documents(documents)\n",
@ -81,7 +81,7 @@
},
{
"cell_type": "code",
"execution_count": 6,
"execution_count": 4,
"metadata": {},
"outputs": [
{
@ -100,7 +100,6 @@
],
"source": [
"import tempfile\n",
"import os\n",
"\n",
"persist_path = os.path.join(tempfile.gettempdir(), \"union.parquet\")\n",
"\n",
@ -185,32 +184,6 @@
"print(docs[0].page_content)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Filter"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"1\n"
]
}
],
"source": [
"_filter = {\"id\": \"c53e6eac-0070-403c-8435-a9e528539610\"}\n",
"docs = vector_store.similarity_search(query, filter=_filter)\n",
"print(len(docs))"
]
},
{
"cell_type": "markdown",
"metadata": {},
@ -244,7 +217,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.1"
"version": "3.10.6"
}
},
"nbformat": 4,

80
libs/langchain/langchain/vectorstores/sklearn.py
View File

@ -233,66 +233,33 @@ class SKLearnVectorStore(VectorStore):
return list(zip(neigh_idxs[0], neigh_dists[0]))
def similarity_search_with_score(
self,
query: str,
*,
k: int = DEFAULT_K,
fetch_k: int = DEFAULT_FETCH_K,
filter: Optional[Dict[str, Any]] = None,
**kwargs: Any,
self, query: str, *, k: int = DEFAULT_K, **kwargs: Any
) -> List[Tuple[Document, float]]:
query_embedding = self._embedding_function.embed_query(query)
indices_dists = self._similarity_index_search_with_score(
query_embedding, k=fetch_k, **kwargs
query_embedding, k=k, **kwargs
)
docs: List[Tuple[Document, float]] = []
for idx, dist in indices_dists:
doc = (
return [
(
Document(
page_content=self._texts[idx],
metadata={"id": self._ids[idx], **self._metadatas[idx]},
),
dist,
)
if filter is None:
docs.append(doc)
else:
filter = {
key: [value] if not isinstance(value, list) else value
for key, value in filter.items()
}
if all(
doc[0].metadata.get(key) in value for key, value in filter.items()
):
docs.append(doc)
return docs[:k]
for idx, dist in indices_dists
]
def similarity_search(
self,
query: str,
k: int = DEFAULT_K,
fetch_k: int = DEFAULT_FETCH_K,
filter: Optional[Dict[str, Any]] = None,
**kwargs: Any,
self, query: str, k: int = DEFAULT_K, **kwargs: Any
) -> List[Document]:
docs_scores = self.similarity_search_with_score(
query, k=k, fetch_k=fetch_k, filter=filter, **kwargs
)
docs_scores = self.similarity_search_with_score(query, k=k, **kwargs)
return [doc for doc, _ in docs_scores]
def _similarity_search_with_relevance_scores(
self,
query: str,
k: int = DEFAULT_K,
fetch_k: int = DEFAULT_FETCH_K,
filter: Optional[Dict[str, Any]] = None,
**kwargs: Any,
self, query: str, k: int = DEFAULT_K, **kwargs: Any
) -> List[Tuple[Document, float]]:
docs_dists = self.similarity_search_with_score(
query, k=k, fetch_k=fetch_k, filter=filter, **kwargs
)
docs_dists = self.similarity_search_with_score(query, k=k, **kwargs)
docs, dists = zip(*docs_dists)
scores = [1 / math.exp(dist) for dist in dists]
return list(zip(list(docs), scores))
@ -303,7 +270,6 @@ class SKLearnVectorStore(VectorStore):
k: int = DEFAULT_K,
fetch_k: int = DEFAULT_FETCH_K,
lambda_mult: float = 0.5,
filter: Optional[Dict[str, Any]] = None,
**kwargs: Any,
) -> List[Document]:
"""Return docs selected using the maximal marginal relevance.
@ -317,7 +283,6 @@ class SKLearnVectorStore(VectorStore):
of diversity among the results with 0 corresponding
to maximum diversity and 1 to minimum diversity.
Defaults to 0.5.
filter: (Optional[Dict[str, str]]): Filter by metadata. Defaults to None.
Returns:
List of Documents selected by maximal marginal relevance.
"""
@ -329,28 +294,17 @@ class SKLearnVectorStore(VectorStore):
mmr_selected = maximal_marginal_relevance(
self._np.array(embedding, dtype=self._np.float32),
result_embeddings,
k=fetch_k,
k=k,
lambda_mult=lambda_mult,
)
mmr_indices = [indices[i] for i in mmr_selected]
docs = []
for idx in mmr_indices:
doc = Document(
return [
Document(
page_content=self._texts[idx],
metadata={"id": self._ids[idx], **self._metadatas[idx]},
)
if filter is None:
docs.append(doc)
else:
filter = {
key: [value] if not isinstance(value, list) else value
for key, value in filter.items()
}
if all(doc.metadata.get(key) in value for key, value in filter.items()):
docs.append(doc)
return docs[:k]
for idx in mmr_indices
]
def max_marginal_relevance_search(
self,
@ -358,7 +312,6 @@ class SKLearnVectorStore(VectorStore):
k: int = DEFAULT_K,
fetch_k: int = DEFAULT_FETCH_K,
lambda_mult: float = 0.5,
filter: Optional[Dict[str, Any]] = None,
**kwargs: Any,
) -> List[Document]:
"""Return docs selected using the maximal marginal relevance.
@ -372,7 +325,6 @@ class SKLearnVectorStore(VectorStore):
of diversity among the results with 0 corresponding
to maximum diversity and 1 to minimum diversity.
Defaults to 0.5.
filter: (Optional[Dict[str, str]]): Filter by metadata. Defaults to None.
Returns:
List of Documents selected by maximal marginal relevance.
"""
@ -383,7 +335,7 @@ class SKLearnVectorStore(VectorStore):
embedding = self._embedding_function.embed_query(query)
docs = self.max_marginal_relevance_search_by_vector(
embedding, k, fetch_k, lambda_mul=lambda_mult, filter=filter, **kwargs
embedding, k, fetch_k, lambda_mul=lambda_mult
)
return docs

40
libs/langchain/tests/unit_tests/vectorstores/test_sklearn.py
View File

@ -12,7 +12,7 @@ def test_sklearn() -> None:
"""Test end to end construction and search."""
texts = ["foo", "bar", "baz"]
docsearch = SKLearnVectorStore.from_texts(texts, FakeEmbeddings())
output = docsearch.similarity_search("foo", k=1, fetch_k=3)
output = docsearch.similarity_search("foo", k=1)
assert len(output) == 1
assert output[0].page_content == "foo"
@ -27,24 +27,10 @@ def test_sklearn_with_metadatas() -> None:
FakeEmbeddings(),
metadatas=metadatas,
)
output = docsearch.similarity_search("foo", k=1, fetch_k=3)
output = docsearch.similarity_search("foo", k=1)
assert output[0].metadata["page"] == "0"
@pytest.mark.requires("numpy", "sklearn")
def test_sklearn_with_metadatas_and_filter() -> None:
"""Test end to end construction and search."""
texts = ["foo", "foo", "bar", "baz"]
metadatas = [{"page": str(i)} for i in range(len(texts))]
docsearch = SKLearnVectorStore.from_texts(
texts,
FakeEmbeddings(),
metadatas=metadatas,
)
output = docsearch.similarity_search("foo", k=1, fetch_k=4, filter={"page": "1"})
assert output[0].metadata["page"] == "1"
@pytest.mark.requires("numpy", "sklearn")
def test_sklearn_with_metadatas_with_scores() -> None:
"""Test end to end construction and scored search."""
@ -55,7 +41,7 @@ def test_sklearn_with_metadatas_with_scores() -> None:
FakeEmbeddings(),
metadatas=metadatas,
)
output = docsearch.similarity_search_with_relevance_scores("foo", k=1, fetch_k=3)
output = docsearch.similarity_search_with_relevance_scores("foo", k=1)
assert len(output) == 1
doc, score = output[0]
assert doc.page_content == "foo"
@ -75,7 +61,7 @@ def test_sklearn_with_persistence(tmpdir: Path) -> None:
serializer="json",
)
output = docsearch.similarity_search("foo", k=1, fetch_k=3)
output = docsearch.similarity_search("foo", k=1)
assert len(output) == 1
assert output[0].page_content == "foo"
@ -85,7 +71,7 @@ def test_sklearn_with_persistence(tmpdir: Path) -> None:
docsearch = SKLearnVectorStore(
FakeEmbeddings(), persist_path=str(persist_path), serializer="json"
)
output = docsearch.similarity_search("foo", k=1, fetch_k=3)
output = docsearch.similarity_search("foo", k=1)
assert len(output) == 1
assert output[0].page_content == "foo"
@ -112,19 +98,3 @@ def test_sklearn_mmr_by_vector() -> None:
)
assert len(output) == 1
assert output[0].page_content == "foo"
@pytest.mark.requires("numpy", "sklearn")
def test_sklearn_mmr_with_metadata_and_filter() -> None:
"""Test end to end construction and search."""
texts = ["foo", "foo", "bar", "baz"]
metadatas = [{"page": str(i)} for i in range(len(texts))]
docsearch = SKLearnVectorStore.from_texts(
texts, FakeEmbeddings(), metadatas=metadatas
)
output = docsearch.max_marginal_relevance_search(
"foo", k=1, fetch_k=4, filter={"page": "1"}
)
assert len(output) == 1
assert output[0].page_content == "foo"
assert output[0].metadata["page"] == "1"