Learning of Piece Values for Chess Variants

In this paper we describe a set of experiments for learning a the values of chess pieces for the popular chess variants crazyhouse, suicide, and atomic chess. We follow an established methodology that relies on reinforcement learning from self-games to learn piece values and piece-square tables for these chess variants. The learned piece values are quite different from those of standard chess, and in several ways surprising, but they generally outperform those that have been previously used in the literature, and in implementations of computer players for these games. The results also underline the practical importance of piece-square value tables for tactical variants of the game.