RuleEnhanced Penalized Regression by Column Generation using Rectangular Maximum Agreement
[edit]
Proceedings of the 34th International Conference on Machine Learning, PMLR 70:10591067, 2017.
Abstract
We describe a learning procedure enhancing L1penalized regression by adding dynamically generated rules describing multidimensional “box” sets. Our ruleadding procedure is based on the classical column generation method for highdimensional linear programming. The pricing problem for our column generation procedure reduces to the NPhard rectangular maximum agreement (RMA) problem of finding a box that best discriminates between two weighted datasets. We solve this problem exactly using a parallel branchandbound procedure. The resulting ruleenhanced regression procedure is computationintensive, but has promising prediction performance.
Related Material


