Question on (testing) cosine_similiarity with WHERE clauses about pgvector-python

[pmalacho-mit]

Great, thanks @ankane ! That's very clear.

I actually had tried that, which was not working for me, but now that you mention it I again I realize it's simply because I was relying on a default_factory function to generate those UUIDs fresh it each time.

I have a feeling that was actually causing the really wacky non-determinism I was seeing, and just didn't realize in my confusion.

Thanks for the quick reply and the help!

[ankane]

Hi @pmalacho-mit, if two rows have the same distance (like in the example above), the database can return them in either order. You can add a second sort expression to make it deterministic.

SELECT ... ORDER BY vector <=> $1, id

https://www.postgresql.org/docs/current/queries-order.html

Edit: With SQLModel, this would be:

order_by(Model.vector.cosine_distance([1, 1, 1]), Model.id)