How to write code for Cold Start scenarios with LightFM library

Question

I don't know how to write code to manage cold start recommendation with LightFM library.

Introduction

Suppose to have the following datasets:

• rating_dataset

userId	movieId	rating
1	1	4
1	3	4
2	318	3

etc...

• movie_features

movieId	genres
1	Adventure, Animation, Children, Comedy, Fantasy
2	Adventure, Children, Fantasy
3	Comedy, Romance

etc...

• user_features

userId	location	age
1	new york	12
2	moscow	37
3	rio de janeiro	20

etc...

There are different scenarios where userId ad movieId may be:

1. both present
2. one new and the other already present (item or user cold start, the new one is not present in it's corresponding feature matrix)
3. both new (item and user cold start, not present in both feature matrices too) N.B. New only ids are new, the features are a combination of those of the corresponding feature matrix.

e.g. of new row

new_id, new york (already present), 20 (already present)

For each combination I may have user_features and item_features or None.

Code

For case 1. I wrote the following code:

    already_known_user_ids = np.array([...])
    already_known_item_ids = np.array([...])

    #with features
    model.predict(already_known_user_ids, already_known_item_ids, item_features, user_features)

    #without features
    model.predict(already_known_user_ids, already_known_item_ids)

No problems with that, I think I wrote it correctly.

For cases 2. and 3. I can't use new ids directly in .predict() function because they aren't present during .fit() of the model. Furthermore, I can't add manually new ids to the corresponding feature matrix and use it in .predict() because I get the error ValueError: User id new_user_id not in user id mapping. Make sure you call the fit method.

    #not working code
    new_user = np.array([new_user_id])
    already_known_item_ids = np.array([...])
    #user_features doesn't contain new_user_id
    model.predict(new_user, already_known_item_ids, item_features, user_features)

Questions

1. How have I to write code for cases 2. and 3.?
2. What happens if new id has unique new feature? e.g. for user_features: new_user_id, new_city, 56 or for movie_features: new_movie_id, new_genre

Answer 1

Not a solution but a workaround: I encountered the very same error when removing Visual Studio 2019 to fully rely on Visual Studio 2022. Reinstalling a leight-weigth version of VS2019 to have access to its VSIXInstaller.exe worked for me.

Note: This seems not to happen on every machine. I tried same setup on a virtual machine and had no need for VS2019. So this might only occur on computers that had formerly an older version of Visual Studio installed.