Analysis of Performance of Consultation Methods in Computer Chess

The performance of consultation methods, i.e., majority voting, the optimistic selection rule, and the pseudo-random number (PRN) ensemble method, are examined in computer chess using 2180 chess problems. Here, the optimistic selection rule selects a program that returns the highest search value, and the PRN ensemble consists of multiple individual copies of one base program, and each copy is diversified by adding random numbers to the evaluation function of the base program. We carried out empirical experiments by using state-of-the-art chess-program Crafty as the base program. We found that the percentage of correct answers increased from 55.6 to 57.1% using optimistic selection from the PRN ensemble. The experimental results indicated that the consultation methods allowed simple yet effective distributed computing in chess.