[Question] Normalizing vectors

Hi there, I’m generating my own vectors. I plan to use PQ compression.
Will it improve search results if I normalize vectors (transform each vector to length of 1) before inserting them into a collection?
Thank you for your time.

hi @cpwalker ! Welcome to our community :hugs:

You mean each vector with only 1 dimension? :thinking:

When configuring PQ, you can set the segments yourself:

So I believe it’s best to just pass the vector with the number of dimensions you get from your model, and let Weaviate compress those dimensions into the number of segments you define.

AFAIK, less segments you set for the compression, less recall you will get for your results.

Let me know if this helps!


Hi Duda, thanks for your reply. No, my vectors don’t have 1 dimension. I mean the computation of a normalized vector, i.e. https://mathworld.wolfram.com/NormalizedVector.html.