Description

ImplicitCF raises an IndexError if the user appears in the test dataset but not on the training dataset.

How do we replicate the issue?

Split a dataset using a method like TimeSeriesSplit or python_chrono_split. I.e: len(ImplicitCF.interact_status) < len(ImplicitCF.user_idx)

Expected behavior (i.e. solution)

Raisign a meaningful error if the dataset needs to be stratified, or assuming that if the user is not on the ImplicitCF.interact_status table, it should have the empty set of items.

Other Comments

Meanwhile, I solved it by using:

data.interact_status = data.interact_status.reindex(data.user_idx['userID_idx'])
data.interact_status['userID'] = data.interact_status.index
data.interact_status['itemID_interacted'] = data.interact_status['itemID_interacted'].fillna("").apply(set)

This will create a the remaining "empty" users

Or just deleting items in test that don't appear in train