Closed anna-charlotte closed 6 months ago
Hi @anna-charlotte, Thanks for highlighting the issue! It seems like we can also rename cosine similarity to cosine distance to avoid this confusion. Changing the formula will break previous code.
Hi @yurymalkov sure sounds good, too 👍
Hello there,
I think there has been a mix up with cosine similarity and cosine distance. In the docs it says that you support
cosine similarity
. But instead the cosine distance is calculated (1- cosine sim.). The resulting order is still correct, but when looking at the scores, they are incorrect. For instance, when computing the distance of vectorx = [1.0, 1.0, 1.0, 0.0]
to itself, the resulting score is 0.0. For the cosine similarity it would be 1. though instead of 0.Could you adjust the formula?