Nicolas Hug

No worries, it can wait. But I'm sure you understand I cannot merge anything that is not thoroughly tested, especially when it's such a big feature / improvement. EDIT: I...

@martincousi is 100% correct No plan to merge this (or the other branch) unfortunately, because I don't have enough visibility on how well it would integrate with the current code...

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

For most prediction algorithms this isn't possible without (at least partially) re-training the models with the new users. Surprise does not support partial fitting although there are some discussions and...

I agree that design for this part could be improved a lot. Could you please give details about the PR? Also all tests seem to be failing?

PR = Pull Request. The question is: what changes are you proposing?

If the changes only affect `Dataset.split()` then it's probably not worth spending too much time on it: the `split()` method is deprecated and replaced by the use of CV iterators.

Given the differences between surprise and scikit-learn, I think the analoguous way of passing the weights into the `fit` method would be for us to include it in the `trainset`...

I'm not completely clear on what the weights are for TBH. Is it just for computing error / accuracy metrics? In this case the changes could be minimal and restrained...

reducing the number of iterations and the number of factors will result in faster training time. But that will potentially hurt RMSE