Selective Search in an Amazons Program

Throughout at least the first two thirds of an Amazons game the number of legal moves is so great that doing a full width search is impractical. This means we must resort to some sort of selective search and in this paper we study a number of algorithms to perform forward pruning. In particular we describe techniques for selecting from the set of all legal moves a small subset of moves for look-ahead evaluation, eliminating all other moves from consideration. We then study the effects of these techniques experimentally and show which have the most potential for producing the best Amazons-playing program.