mirror of
https://github.com/Mintplex-Labs/langchain-python.git
synced 2026-07-21 00:35:23 -04:00
f907b62526
# Scores in Vectorestores' Docs Are Explained Following vectorestores can return scores with similar documents by using `similarity_search_with_score`: - chroma - docarray_hnsw - docarray_in_memory - faiss - myscale - qdrant - supabase - vectara - weaviate However, in documents, these scores were either not explained at all or explained in a way that could lead to misunderstandings (e.g., FAISS). For instance in FAISS document: if we consider the score returned by the function as a similarity score, we understand that a document returning a higher score is more similar to the source document. However, since the scores returned by the function are distance scores, we should understand that smaller scores correspond to more similar documents. For the libraries other than Vectara, I wrote the scores they use by investigating from the source libraries. Since I couldn't be certain about the score metric used by Vectara, I didn't make any changes in its documentation. The links mentioned in Vectara's documentation became broken due to updates, so I replaced them with working ones. VectorStores / Retrievers / Memory - @dev2049 my twitter: [berkedilekoglu](https://twitter.com/berkedilekoglu) --------- Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
202 lines
6.7 KiB
Python
202 lines
6.7 KiB
Python
from abc import ABC
|
|
from typing import TYPE_CHECKING, Any, Iterable, List, Optional, Tuple, Type
|
|
|
|
import numpy as np
|
|
from pydantic import Field
|
|
|
|
from langchain.embeddings.base import Embeddings
|
|
from langchain.schema import Document
|
|
from langchain.vectorstores import VectorStore
|
|
from langchain.vectorstores.utils import maximal_marginal_relevance
|
|
|
|
if TYPE_CHECKING:
|
|
from docarray import BaseDoc
|
|
from docarray.index.abstract import BaseDocIndex
|
|
|
|
|
|
def _check_docarray_import() -> None:
|
|
try:
|
|
import docarray
|
|
|
|
da_version = docarray.__version__.split(".")
|
|
if int(da_version[0]) == 0 and int(da_version[1]) <= 31:
|
|
raise ValueError(
|
|
f"To use the DocArrayHnswSearch VectorStore the docarray "
|
|
f"version >=0.32.0 is expected, received: {docarray.__version__}."
|
|
f"To upgrade, please run: `pip install -U docarray`."
|
|
)
|
|
except ImportError:
|
|
raise ImportError(
|
|
"Could not import docarray python package. "
|
|
'Please install it with `pip install "langchain[docarray]"`.'
|
|
)
|
|
|
|
|
|
class DocArrayIndex(VectorStore, ABC):
|
|
def __init__(
|
|
self,
|
|
doc_index: "BaseDocIndex",
|
|
embedding: Embeddings,
|
|
):
|
|
"""Initialize a vector store from DocArray's DocIndex."""
|
|
self.doc_index = doc_index
|
|
self.embedding = embedding
|
|
|
|
@staticmethod
|
|
def _get_doc_cls(**embeddings_params: Any) -> Type["BaseDoc"]:
|
|
"""Get docarray Document class describing the schema of DocIndex."""
|
|
from docarray import BaseDoc
|
|
from docarray.typing import NdArray
|
|
|
|
class DocArrayDoc(BaseDoc):
|
|
text: Optional[str]
|
|
embedding: Optional[NdArray] = Field(**embeddings_params)
|
|
metadata: Optional[dict]
|
|
|
|
return DocArrayDoc
|
|
|
|
@property
|
|
def doc_cls(self) -> Type["BaseDoc"]:
|
|
if self.doc_index._schema is None:
|
|
raise ValueError("doc_index expected to have non-null _schema attribute.")
|
|
return self.doc_index._schema
|
|
|
|
def add_texts(
|
|
self,
|
|
texts: Iterable[str],
|
|
metadatas: Optional[List[dict]] = None,
|
|
**kwargs: Any,
|
|
) -> List[str]:
|
|
"""Run more texts through the embeddings and add to the vectorstore.
|
|
|
|
Args:
|
|
texts: Iterable of strings to add to the vectorstore.
|
|
metadatas: Optional list of metadatas associated with the texts.
|
|
|
|
Returns:
|
|
List of ids from adding the texts into the vectorstore.
|
|
"""
|
|
ids: List[str] = []
|
|
embeddings = self.embedding.embed_documents(list(texts))
|
|
for i, (t, e) in enumerate(zip(texts, embeddings)):
|
|
m = metadatas[i] if metadatas else {}
|
|
doc = self.doc_cls(text=t, embedding=e, metadata=m)
|
|
self.doc_index.index([doc])
|
|
ids.append(str(doc.id))
|
|
|
|
return ids
|
|
|
|
def similarity_search_with_score(
|
|
self, query: str, k: int = 4, **kwargs: Any
|
|
) -> List[Tuple[Document, float]]:
|
|
"""Return docs most similar to query.
|
|
|
|
Args:
|
|
query: Text to look up documents similar to.
|
|
k: Number of Documents to return. Defaults to 4.
|
|
|
|
Returns:
|
|
List of documents most similar to the query text and
|
|
cosine distance in float for each.
|
|
Lower score represents more similarity.
|
|
"""
|
|
query_embedding = self.embedding.embed_query(query)
|
|
query_doc = self.doc_cls(embedding=query_embedding) # type: ignore
|
|
docs, scores = self.doc_index.find(query_doc, search_field="embedding", limit=k)
|
|
|
|
result = [
|
|
(Document(page_content=doc.text, metadata=doc.metadata), score)
|
|
for doc, score in zip(docs, scores)
|
|
]
|
|
return result
|
|
|
|
def similarity_search(
|
|
self, query: str, k: int = 4, **kwargs: Any
|
|
) -> List[Document]:
|
|
"""Return docs most similar to query.
|
|
|
|
Args:
|
|
query: Text to look up documents similar to.
|
|
k: Number of Documents to return. Defaults to 4.
|
|
|
|
Returns:
|
|
List of Documents most similar to the query.
|
|
"""
|
|
results = self.similarity_search_with_score(query, k=k, **kwargs)
|
|
return [doc for doc, _ in results]
|
|
|
|
def _similarity_search_with_relevance_scores(
|
|
self,
|
|
query: str,
|
|
k: int = 4,
|
|
**kwargs: Any,
|
|
) -> List[Tuple[Document, float]]:
|
|
"""Return docs and relevance scores, normalized on a scale from 0 to 1.
|
|
|
|
0 is dissimilar, 1 is most similar.
|
|
"""
|
|
raise NotImplementedError
|
|
|
|
def similarity_search_by_vector(
|
|
self, embedding: List[float], k: int = 4, **kwargs: Any
|
|
) -> List[Document]:
|
|
"""Return docs most similar to embedding vector.
|
|
|
|
Args:
|
|
embedding: Embedding to look up documents similar to.
|
|
k: Number of Documents to return. Defaults to 4.
|
|
|
|
Returns:
|
|
List of Documents most similar to the query vector.
|
|
"""
|
|
|
|
query_doc = self.doc_cls(embedding=embedding) # type: ignore
|
|
docs = self.doc_index.find(
|
|
query_doc, search_field="embedding", limit=k
|
|
).documents
|
|
|
|
result = [
|
|
Document(page_content=doc.text, metadata=doc.metadata) for doc in docs
|
|
]
|
|
return result
|
|
|
|
def max_marginal_relevance_search(
|
|
self,
|
|
query: str,
|
|
k: int = 4,
|
|
fetch_k: int = 20,
|
|
lambda_mult: float = 0.5,
|
|
**kwargs: Any,
|
|
) -> List[Document]:
|
|
"""Return docs selected using the maximal marginal relevance.
|
|
|
|
Maximal marginal relevance optimizes for similarity to query AND diversity
|
|
among selected documents.
|
|
|
|
Args:
|
|
query: Text to look up documents similar to.
|
|
k: Number of Documents to return. Defaults to 4.
|
|
fetch_k: Number of Documents to fetch to pass to MMR algorithm.
|
|
lambda_mult: Number between 0 and 1 that determines the degree
|
|
of diversity among the results with 0 corresponding
|
|
to maximum diversity and 1 to minimum diversity.
|
|
Defaults to 0.5.
|
|
Returns:
|
|
List of Documents selected by maximal marginal relevance.
|
|
"""
|
|
query_embedding = self.embedding.embed_query(query)
|
|
query_doc = self.doc_cls(embedding=query_embedding) # type: ignore
|
|
|
|
docs = self.doc_index.find(
|
|
query_doc, search_field="embedding", limit=fetch_k
|
|
).documents
|
|
|
|
mmr_selected = maximal_marginal_relevance(
|
|
np.array(query_embedding), docs.embedding, k=k
|
|
)
|
|
results = [
|
|
Document(page_content=docs[idx].text, metadata=docs[idx].metadata)
|
|
for idx in mmr_selected
|
|
]
|
|
return results
|