Algorithms don't return more than a certain number of recommendations per user

[abdollahpouri]

I was trying to get a recommendation of size 100 per user for MovieLens 1M dataset but the algorithm(I tried SVD but I'm sure it is the same for others too) does not recommend that many items for all users (for some users only recommend 10 or maybe a bit more).

Is there any way we can set how many recommendations we want regardless of whether they have high predictions or not? For example for each uid, return 100 iid and their scores even if the scores are low it id fine. We need this capability since many re-ranker algorithms (diversity, calibration etc. ) need a large list generated by a standard algorithm (100 or larger) to build the final list, and right now this is not possible to do on Surprise.

[rafmesal]

I have the same problem, I need 5 recommendatios for every user, but SVD returns between 1 - 5 recommendations.

[NicolasHug]

Hi, have you tried https://surprise.readthedocs.io/en/stable/FAQ.html#how-to-get-the-top-n-recommendations-for-each-user ?

I'm getting 100 recommendations for all users on MovieLens1M and SVD

[chiangandy]

Hi, have you tried https://surprise.readthedocs.io/en/stable/FAQ.html#how-to-get-the-top-n-recommendations-for-each-user ?

I'm getting 100 recommendations for all users on MovieLens1M and SVD

when I try this, I found it take so long time in trainset.build_anti_testset()(I have no GPU env, so pure CPU env running).
Also, according to algorithm, we need to find all of new item set by trainset.build_anti_testset() then try each set in predictions and sort it to get optimizing result, which will take terrible long time! In my opinion I think it is impossible to use in online function.
As a result, I think this package is not suitable to get top N recommendation by user.

Sorry, just pending to use this.