Learning Evaluation Functions to Improve Local Search

This paper describes Stage, a learning algorithm that automatically improves search performance on large-scale optimization problems. Stage learns an evaluation function that predicts the outcome of a local search algorithm, such as hillclimbing or Walksat, from features of states visited during search. The learned evaluation function is used to bias future search trajectories toward better optima on the same problem. This paper presents the Stage algorithm; an extension, X-Stage, that transfers learned evaluation functions to new, similar optimization problems; and empirical results on seven large-scale optimization domains: bin-packing, channel routing, Bayes network structure-finding, radiotherapy treatment planning, cartogram design, Boolean satisfiability, and Boggle board setup.