Surprise
Surprise copied to clipboard
NicolasHug
Add adjusted cosine similarity
As discussed in #135, it would be interesting to add support for the adjusted cosine similarity.
Adjusted cosine is similar to pearson similarity, only the centering differs (for a user-user similarity, we center with the item average instead of with the respective user averages).
Yes it's still open so you're welcome to try if you want to. Don't forget to check the contributing guidelines :)
Ok that is cool.
But it seems like there is already an implementation for it following the discussion in the issue #135 ??!!
There's no PR related to adjusted cosine so you can go for it. One of the open PRs is related #164, but even then, it's only partially addressed.
Just to make it clear: implementing adjusted cosine should be done here only considering common ratings. I am aware that adjusted cosine is rather commonly computed on the whole ratings, but that can be dealt with once we address #164.
Ok. Just to get it right, this would be similar to the actual implementation of cosine similiraty with common_ratings_only=True but with the shift at the mean right ?
this would be similar to the actual implementation of cosine similiraty
Yes, but adjusted cosine is even closer to pearson similarity.
with common_ratings_only=True
There is no common_ratings_only option for now, because similarities are always computed using common ratings. This option will come once we address #164.
Ok. I'll try to submit a PR ASAP.
Le lun. 30 avr. 2018 à 13:41, Nicolas Hug [email protected] a écrit :
this would be similar to the actual implementation of cosine similiraty
Yes, but adjusted cosine is even closer to pearson similarity.
with common_ratings_only=True
There is no common_ratings_only option for now, because similarities are always computed using common ratings. This option will come once we address #164 https://github.com/NicolasHug/Surprise/issues/164.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NicolasHug/Surprise/issues/163#issuecomment-385374643, or mute the thread https://github.com/notifications/unsubscribe-auth/ATl3ioOyVPFd8XiYD8N3RtJ3LytvUnKmks5ttvh9gaJpZM4TFU5W .
hey there I'm new here, Is there any one to help me? I want to add a new similarity measure but there are already some bugs in the similarities.pyx how can I use it? how can I fix it?
@ghazalaak please don't highjack unrelated issues. Open a new one if you need to (you'll have to be a lot more specific though).