Reducing Nearest Neighbor Training Sets Optimally and Exactly