Prior and Likelihood Choices for Bayesian Matrix Factorisation on Small Datasets