lyst
Cannot run predictions in batches
Hi ,
I am trying now to predict for batch users, so I am using your suggestion from issue #240.
I am getting an error after first batch.
Do I understand correctly, that in liпhtfm.py on the row 814: n_users = user_ids.max() + 1 , meaning that if I will send users_ids for the first batch, it will work, but starting from second not, because user_ids=[7,7,7,8,8,8] and features [1,2,3,1,2,3] will give n_users=9 and user_features.shape[0]=6 and throught an error "Number of user feature rows does not equal the number of users"
What am doing wrong?
Originally posted by @maffka123 in https://github.com/lyst/lightfm/issues/240#issuecomment-878569973
I am a novice at LightFM so I advise you to look at the code yourself about what I am saying. But for your question:
Within predict() method, the parameter "user_ids" we use as the representation of users isn't the same user IDs we have fitted our model with, the predict parameter IDs does not point to our initial user_ids list, they are independent of each other. Instead, these ids within predict method work as the indices of user_features within the predict method parameters.
In other words when you try to use user_ids=[7,7,7,8,8,8], the predict method tries to find the 8th indexed element of user_features you put inside predict parameters which doesn't exist. Instead of that, you should simply change [7,7,7,8,8,8] into [0,0,0,1,1,1]. This wouldn't cause a problem since user_ids in this context simply point at user_features[0] or user_features[1] depending on their ids.
This fact essentially let's us predict the recommendations of unseen users with the help of their known features without having to add them into our datasets.
@ilkeremrekoc Thanks a lot, I was getting the same error. Now it works!!
