Robust Constraint Solving Using Multiple Heuristics

Abstract. Constraint Programming is a proven successful technique, but it requires skill in modeling problems, and knowledge on how algorithms interact with models. What can be a good algorithm for one problem class can be very poor for another; even within the same class performance can vary wildly from one instance to another. CP could be easier to use if we could design robust algorithms that perform well across a range of problems, models and instances. In this paper we look specifically at variable and value ordering heuristics for backtracking search and propose a multi-heuristic algorithm based on time-slicing, and we demonstrate its performance on two different problem classes, showing it is more robust than the standard heuristics.

• Type: Conference
• Year: 2005
• Where: CP
• Authors: Alfio Vidotto, Kenneth N. Brown, J. Christopher Beck
• Keywords: Algorithm, CP 2005, Programming Languages, Proven Successful Technique, Value Ordering Heuristics