Training model with SVD++ on the Movie Lens 100k dataset