Classifier-based constraint acquisition

Abstract Modeling a combinatorial problem is hard and error-prone task requiring significant expertise. Constraint acquisition methods attempt to automate this process by learning constraints from examples of solutions (usually) non-solutions. Active query an oracle while passive do not. We propose known but not widely-used application machine constraint acquisition: training classifier discriminate between non-solutions, then deriving model the trained classifier. discuss wide range possible new with useful properties inherited classifiers. also show potential approach using Naive Bayes classifier, obtaining algorithm that considerably faster than existing methods, scalable large sets, robust under errors.