Closed danielfriis closed 7 months ago
Hi @danielfriis, the nearest neighbor search should have 100% accuracy if there's not an approximate index. I'd double check the query is correct (if you have a default_scope
on the model, it'll take precedence until 0.4.0) as well as the distance function (you'll typically want to use cosine
). For general best practices for RAG, it'd be better to use other resources.
Edit: You may also want to try different embedding models and chunking strategies to see if it makes a difference.
@ankane thanks a lot. That makes 100p sense!
This might not be the right place to ask this question, in which case, feel free to close the issue!
I'm using neighbor to build a RAG system using a vector db (pgvector).
I have a bunch of very different documents where I need to find an answer, but I struggle to retrieve the correct chunk.
Are there any best practices for making the results better?