How to use a collab_learner to get predictions?

After building a collab_learner model, how exactly does one use it?
That is, assume there is a user (not from the validation or training set) for whom we have the rating of 30 movies, how exactly do we use the user’s data to generate a movie suggestion for them?

Hi M
Have a look at page 267 and 270 of the book.
If your users really like film A and film B then you can find films with a similar embedding distance. If you are interested in Principle Component Analysis it is covered in Rachel’s Computational Numeric Analysis course.
Regards Conwyn

1- So, there is no way of finding the embedding vector of this user and multiplying it with the embedding matrice of the movies? This is not what recommendation engines would typically do?

2- “If your users really like film A and film B then you can find films with a similar embedding distance.” In the case that the user likes a set of N movies, do we then look for the set of unwatched 5 movies that have the minimum mean distance to the user set of N-rated movies?
I can pinpoint a few tedious issues here: the user usually rates the movies on a scale, rather than a 0/1 rating. So, one should find movies that are far away from 1-star rated movies and close to 5-star rated movies but, what about the values in between? Should we scale the distances by some factor? What is it? All of this feels like an optimization problem that should be handled in ML fashion, right?

Again my question is, what is missing to just find the embedding vector for this user?