Is the L2 distance score in vector search an inverted score? #1261
-
Hi there, the L2 scores and cosine scores for LanceDB vector search suggests that smaller |
Beta Was this translation helpful? Give feedback.
Answered by
westonpace
May 6, 2024
Replies: 1 comment
-
The L2 score is not inverted. The cosine score is "cosine distance" and not "cosine similarity". Cosine distance is defined as In all cases a smaller |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
yy0vnne
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The L2 score is not inverted.
The cosine score is "cosine distance" and not "cosine similarity".
Cosine distance is defined as
1 - cosine_similarity
(cosine distance is inverted cosine similarity).In all cases a smaller
_distance
means more similar.