Using Racing to Automatically Configure Algorithms for Scaling Performance

Automated algorithm configuration has been proven to be an effective approach for achieving improved performance of solvers for many computationally hard problems. Following our previous work, we consider the challenging situation where the kind of problem instances for which we desire optimised performance are too difficult to be used during the configuration process. In this work, we propose a novel combination of racing techniques with existing algorithm configurators to meet this challenge. We demonstrate that the resulting algorithm configuration protocol achieves better results than previous approaches and in many cases closely matches the bound on performance obtained using an oracle selector. An extended version of this paper can be found at www.cs.ubc.ca/labs/beta/Projects/Config4Scaling .