Fix one set of latent representations?

[phiweger]

I am new to recommender systems and would like to use latent representations for users as documented here:

https://maciejkula.github.io/spotlight/factorization/representations.html#module-spotlight.factorization.representations

I already trained representations (an embedding) for the items though, and would like to keep them fix, i.e. freeze them during training -- Is this possible? If so, how should I implement this?

Thank you very much.

[maciejkula]

You could try writing your own representation module, and fix the embeddings there.