Hello community,
I would like to share a package built with pytorch to train recommendation system models https://github.com/amoussawi/recoder.
It contains two factorization models implementations: (Deep) Autoencoders and Matrix Factorization. You can build your own factorization model and train it.
It was built to be fast for large-scale training with negative sampling, for instance you can have an Autoencoder model fully trained on Movielens-20M dataset in less than a minute on a Tesla K80 GPU.