Multi-value Rule Sets for Interpretable Classification with Feature-Efficient Representations