I am working through the Collaborative filtering part of the fastbook: fastbook/08_collab.ipynb at master · fastai/fastbook · GitHub and I am trying to understand how exactly each batch works. Based on the information, each batch has two columns, one for users and one for movies. So I played around with it a bit and saw that the first example (probably for you something else) was user 661 and movie 659. But in the ratings df, such a combination does not exist. So my question is how exactly are the users and movies combined? What happens with such cases where the combination does not exist?
Answering my own question: The empty record (execution cell number 28) occurs because of the definition of the dataloaders.
dls = CollabDataLoaders.from_df(ratings, item_name='title', bs=64)
The item name does not correspond to the number of the movie but to the title of the movie and my hypothesis is that fastai automatically assigns some number to each movie title. If we change this to:
dls = CollabDataLoaders.from_df(ratings, bs=64)
then the dls.show_batch() will return an x with a valid user_id, movie_id pair.