vectorstore.as_retriever(...) how to query with embedding instead of str ?

[NikhielRahulSingh]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

faiss_retriever = faiss_vector_store.as_retriever(search_kwargs={"k": 10})

How do I query using a vector ?

Description

I want to query using a vector instead of str

System Info

python

[ccurme]

You can use the similarity_search_by_vector method of the underlying vector store.

from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings()
faiss_vector_store = FAISS.from_texts(["apple", "banana"], embeddings)

vector = embeddings.embed_query("juice")


faiss_vector_store.similarity_search_by_vector(vector)

[Document(metadata={}, page_content='apple'),
 Document(metadata={}, page_content='banana')]

If you need the runnable interface (like with the retriever), you can make a short wrapper:

from langchain_core.documents import Document
from langchain_core.runnables import chain


@chain
def search_by_vector(vector: list[float]) -> list[Document]:
    return faiss_vector_store.similarity_search_by_vector(vector)


search_by_vector.invoke(vector)

[NikhielRahulSingh]

thank you for your response, I am using an ensemble retriever :

BM25

bm25_retriever = BM25Retriever.from_texts(
contexts, metadatas=metadatas
)

FAISS

text_embedding_pairs = zip(contexts, context_embeddings)
faiss_vector_store = FAISS.from_embeddings(text_embedding_pairs,context_embeddings)
faiss_retriever = faiss_vector_store.as_retriever(search_kwargs={"k": 10})

Ensemble

ensemble_retriever = EnsembleRetriever(
retrievers=[bm25_retriever,faiss_retriever],
weights=[0.5, 0.5],
)

@chain
def search_by_vector(vector: list[float]) -> list[Document]:
return ensemble_retriever.similarity_search_by_vector(vector)

x = search_by_vector.invoke(query_embeddings[0])

error

AttributeError: 'EnsembleRetriever' object has no attribute 'similarity_search_by_vector'

Is there a way for me to specify which retriever should use the embeddings and which should use the string(BM25)