Logo-amall

Hi We have an application which makes a lot of use of vector representations of images and text. As part of our development aims I am evaluating vector databases. We have a slightly different need - where we have a vector we are not so much interested in the most similar vector, we need to know the most *disimilar* - so it's furthest neighbour rather than nearest neighbour - how would qdrant cope with this ? thanks !

Last active 16 days ago

4 replies

0 views

  • AP

    Hi We have an application which makes a lot of use of vector representations of images and text. As part of our development aims I am evaluating vector databases. We have a slightly different need - where we have a vector we are not so much interested in the most similar vector, we need to know the most disimilar - so it's furthest neighbour rather than nearest neighbour - how would qdrant cope with this ? thanks !

  • KA

    Hi, with dot production distance or cosine you can query with the negated vector. That should give you the vectors with the highest distance to the query.

  • AP

    yeah I was wondering about negated vectors - so is there any limit on the number of vectors returned for such a request ?

  • KA

    There is no such limit, but how many vectors would you expect?

Last active 16 days ago

4 replies

0 views