Learning Qualitatively Diverse and Interpretable Rules for Classification