An evolutionary algorithm for tuning a chess evaluation function

This paper proposes a method for tuning the weights of the evaluation function of a chess program whose search engine is based on evolutionary programming. In our proposed approach, each individual in the population of the evolutionary algorithm represents a virtual player with specific weights of its evaluation function. This differs from most of the previous approaches reported in the literature, in which normally a tournament between virtual players is held, and the final result (win, loss or draw) is used to decide which players will pass to the following generation. The selection mechanism of our proposed algorithm uses games from chess grandmasters to decide which virtual player will pass to the following generation. Our results indicate that the weight values obtained by our approach are similar to the values known from chess theory. Additionally, the standard deviation from the different runs performed, are lower than those reported by authors of previous related approaches.