Performance Analysis of Consultation Methods in Computer Chess

Abstract

The performance of consultation methods is examined in computer chess using two kinds of experiments. One is playing self-play games to observe the winning rates, and the other is solving a collection of chess problems to observe the percentages of the correct answers. It is shown that the winning rate of the optimistic selection rule with 4 base programs against the original one is 61%. Moreover, it is shown that the rate of correct answers with the nominal depth 8 increases from 59% to 70% using the optimistic selection rule with 16 base programs. These results indicate that consultation methods allow us simple yet effective distributed computing in chess.