1373b284bc381890049e92d324f56de0-AuthorFeedback.pdf

ToReviewer#1: LP-baseddataselection: For Figure 1, we indeed used random sampling to select asubset prior to1 applying the LP-based data selection on the reduced dataset. Figure 1 benchmarks the12 performance of SVM1-ODT on medium-sized datasets and shows that our ODT formulation outperforms CART,13 OCT-H, Fair and BinOCT in terms of test accuracywithout using data selection. Hence theauthors48 employ special tools for the specific decision tree, they cannot be generalized to our general tree; for example, [2]49 proposes amaxflow-based MIPformulation.