Experimental Analysis of the Probability Method

Abstract

The Backward Induction Method, which is the most basic algorithm used for game tree searches, has two weak points. First, the move selected by this method is assured to be the best move as far as the search depth of the game tree is concerned, but is not necessarily the best move towards the end of the game. Secondly, the values evaluated for the leaf nodes do not necessarily give the best advantage at the end of the game.In a previous paper, we proposed a new algorithm, the Probability Method, which is useful for games finishing at the constant moves such as Othello.In this paper we compare the Probability Method with the Backward Induction and Bayesian Methods using Othello.Moreover we propose a pruning procedure for the Probability Method and compare it with the alpha-beta pruning procedure used in the Backward Induction Method.We show that the Probability Method is more effective than both of the Backward Induction and Bayesian Methods and that the pruning procedure for the Probability Method is more advantageous than alpha-beta pruning in the Backward Induction Method for some phases.