This code is Python implementation of the paper "A probabilistic framework to incorporate mixed-data type features: Matrix factorization with multimodal side information".
The code requires Numpy and Scipy packages.
The model performs inference for the user and item latent variables. The inputs are sparse rating matrix, multivariate Gaussian side information matrix one for the users and one for the items, and categorical side information matrix (can include more than one categorical feature) one for the users and one for the items. Variational EM is used to perform inference.
Movielens 100K dataset is provided for demonstration.