UCT algorithm in Amazons human-computer games

The main difficulty of Amazons games is the huge branch factor. Traditional NegaMax Search and Alpha-Beta Search can only search a few layers and can´t improve the games effectively. Monte-Carlo Algorithm can improve the games but it needs a huge amount of computing. This paper studies applying UCT algorithm to the Amazons games to overcome this problem in order to balance the search efficiency and computing load. In UCT mini-max tree search, the algorithm select tree node according to the node´s UCB (Upper Confidence Bound) value. Then evaluate the selected node and return the optimal moves. Using C++ programming language, we implemented Amazons human-computer games software. The experiments show that UCT algorithm can implement the search work in Amazons human-computer games and get satisfactory search efficiency.