Performance of a Comprehensive and Efficient Constraint Library Based on Local Search

Constraint satisfaction is one of the major areas in AI that has important real-life applications. Lee et al. propose E-GENET, a stochastic solver for general constraint solving based on iterative repair. Performance figures show that E-GENET compares favorably against tree-search based solvers in many hard problems. On the other hand, global constraints have been shown to be very effective in modeling complicated CSP’s. They have also improved substantially the efficiency of tree-search based solvers in solving real-life problems. In this paper, we present a comprehensive and efficient library of elementary and global constraints for E-GENET. Such a library is essential for applying EGENET to complex real-life applications. We first present the improved performance of some constraints that appear in our previous papers, followed by the implementation details of three additional global constraints available in the CHIP constraint language. Experimental results, using standard benchmarks and a real-life problem, confirm empirically that the E-GENET architecture is comparable to, if not better than, state of the art in constraint solver technology.