Allow float32 matrices to be used as input about implicit HOT 7 CLOSED

Just found a solution,
Input matrix should be float64 type only,
float32 gives this error.

from implicit.

There is some code to specify if you want to generate 32 or 64 bit factors as the output of this function, but the input data needs to be a float64 confidence matrix right now =(

I'll add something soon to allow float32 inputs (and will update this issue when done) - but in the meantime pass in a float64 matrix as a workaround.

from implicit.

Ok, thanks, I am continuing to play with your great library and comparing results to the raw matrix factorization (with missed data points). Is the github is convenient/appropriate place to ask you about different results ?
Also I would like to suggest a small addition: should be an option to initialize X,Y feature vectors before running matrix factorization (to be able to update existing trained model or use other methods for initialization).

P.S. Also thank you for a insightful blog posts with visualisations 👍

from implicit.

Thanks for the feedback!

Either GitHub or email works for contacting me - if you think that it might be of interest to other people, maybe just create another issue, otherwise my email address is on my github profile.

I'll make sure to add the option to train on pre-initialized factors. I'm changing around a little bit anyways (moving to a class with methods for fitting the model, predicting for a user etc) - and I can easily put that in at the same time.

from implicit.

The last commit allows training on pre-initialized factors: 8c18f16#diff-604d1b48a6ae71b2cc39b27249942f12R60. It will only initialize the factors if they haven't been set already.

from implicit.

Thanks,
I am already made modifications to my local copy of repository to be able to supply preinitialised vectors. I see you are refactoring your code to make it more scikit-learn style.

from implicit.

This commit should enable this: #62

from implicit.