Efficient learning of large sets of locally optimal classification rules