Custom distance metric for binary vectors?

Support
1

I have multiple vectors stored in Pinecone that are binary. When defining the query vector, I want to return the vectors in order of the closest Hamming distance but based only on the indices in the query vector that are set to 1. As a simple example:

Query vector: [0, 1, 1, 0, 1]
DB vectors: [[0, 1, 0, 1, 0], [0, 1, 1, 1, 0], [1, 1, 1, 0, 1]]

These DB vectors should be ordered:

  1. [1, 1, 1, 0, 1] (1s in the 1st, 2nd, and 4th index all match the locations of 1s in query vector)
  2. [0, 1, 1, 1, 0] (matches the 1s in the 1st and 2nd index, not the 4th)
  3. [0, 1, 0, 1, 0] (only matches 1 in the 1st index)

Can I use existing metrics in Pinecone to accomplish this? If not, will it be possible to define a custom metric in Pinecone?

2

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.

3

Hi @alex.l.kong. Pinecone supports euclidean, cosine, and dot-product distance metrics today. We might offer additional ones in the future, but it’s not currently on our roadmap to do so.