docs/integrations/retrievers/pinecone_hybrid_search/

[giscus[bot]]

docs/integrations/retrievers/pinecone_hybrid_search/

Pinecone is a vector database with broad functionality.

https://python.langchain.com/docs/integrations/retrievers/pinecone_hybrid_search/

[rferrazd]

1. Does the method .add_texts, works exactly like the update or upsert method from pinecone? If I rerun my script and add the same documents to pinecone, will it just add copy documents or overwrite it?

2. I am using this hybrid retriever for my RAG application, should I fit the bm25 in all my documents chunks in my pinecone index?
   bm25_encoder.fit(corpus)
   where corpus would = all my documents added to pinecone via the retriever.add_texts() method

[feijoes]

@rferrazd

1. As stated in the documentation, the .add_texts() method is optimized using the upsert mechanism.

2. Regarding your question: if the document chunks are extremely large, it could potentially cause issues. However, under normal circumstances, I believe it's fine to proceed with that approach.

[rferrazd]

Can I use the bm25 without fitting it or will it not work then?

[feijoes]

I think you can. But why wouldn't you fit it? It might work, but perhaps not as expected.