A Bayesian Framework for Learning Rule Sets for Interpretable Classification