I have a question about creating a collaborative filtering model with fastai
Must our data only consist of three columns?
What do we do in the case where we have multiple items to recommend … let’s say movies and music
Looking at the picture below
can that data loader have all the columns from the dataset or its just the first three?
I am just a begginer …
hi, have a look at the docs,
https://docs.fast.ai/collab.html#CollabDataBunch.from_df
you’ll see that if your user/item/rating cols are not in the first 3 cols then you can pass it the column names.
when you have multiple types of items to recommend it makes no difference. you can create a model which just does “movies” or “music”, but if you were amazon then it would do “products” and collab doesn’t care what they are.
just bear in mind that if your data is too sparse it’s going to struggle, ie: if 95% of the ratings you have from me are music, it’ll be more difficult to make movie recommendations to me. given enough training data though it’ll still work.
Thank you so much
