Better Initialization Heuristics for Order-based Bayesian Network Structure Learning

Walter Perez Urcia, Denis Deratani Mauá

Abstract: An effective approach for learning Bayesian network structures is to perform a greedy search on the space of variable orderings using a restricted space of parent sets. Typically, the search is initialized with a randomly generated ordering. This can lead to poor local optima and hurt the performance of the method. In this article we develop informed heuristics for generating initial solutions to order-based structure learning search. Experiments with a large collection of real-world data sets demonstrate that our heuristics increase the quality of the solutions found with a negligible overhead.