One-class constraint acquisition with local search
Daniel Sroka, Tomasz P. Pawlak · Proceedings of the Genetic and Evolutionary Computation Conference · 2018
We propose One-Class Constraint Acquisition with Local Search (OCCALS), a novel method for computer-assisted acquisition of Mixed-Integer Linear Programming (MILP) models from examples. OCCALS is designed to help human experts in preparation of MILP models for their systems. OCCALS supports building MILP models from the examples of positive class only, thus requiring relatively cheap to acquire training set, e.g., by observing historical execution of a system. OCCALS effectively handles multimodal distribution of the training set that may happen in practice. We show experimentally the superiority of OCCALS to a state-of-the-art method.