Is there a way to use exact KNN in search?

Hi there! I am currently evaluating pineconeDB as a possible solution for vector storage.

I have successfully pushed ~15k vectors(cosine, 1536 dim, OpenAI embeddings) to pinecone using the python sdk. However I am not seeing results that are as accurate as I would like when compared to elasticsearch’s vector search, most likely due to pinecone using the ANN search algorithm instead of KNN: Ludicrous BERT Search Speeds | Pinecone

Is there a way to use exact KNN on pinecone? I did not see any option to set the search algorithm in the python sdk references or on index creation.

If not, this would be useful for users with smaller datasets! Seems weaviate may already do something like this: Filtered Vector Search | Weaviate - vector database

1 Like

Any luck. I am facing exactly same issue. retrieved docs are not as good as Azure AI search