MongoEngine/mongoengine

using vector search with mongoengine

alexisdal opened this issue · 2 comments

alexisdal commented

hello

I have been using mongoengine on my latest project and I'm happy with it.

somehow, I would like to push vectors (embeddings) into my mongodb collection and perform vector searches. It's apparently a fresh development of mongodb. source => https://www.mongodb.com/products/platform/atlas-vector-search

I made a simple lookup in the documentation and issues of mongoengine but did not find any matches.

what I want to achieve

  1. push vectors in a collection
  2. perform similarity search between a vector and the entire collection, retrieve the K-th closest matches (typically using cosine distance)
  3. using the mongoengine abstraction layer

Any clue about how to do this ?

bagerard commented

There is no "specific" support built in MongoEngine but I'm not sure it's worth having one. There is an API for creating indexes in pymongo raw format, and an api for running aggregation pipeline also in raw format, have you tried to make use of these?

idoshr commented

From what I know about the vector search is part of the elastic search solution that Atlas provide.
Maybe this library can be helpful
https://github.com/certego/AtlasQ