Partial Board Tree Search

Abstract

In this article, a new approach to solving selected two-person perfect information zero-sum games is proposed. This new approach involves dividing game board into smaller parts called partial boards. Every such partial board has an associated “border state” which is an information structure vital to the game solving process. Using partial boards, it is possible to generate partial board game trees. Such structures are sufficient for the purpose of searching complete game trees - in effect. The possibility of making different board divisions allows an entire partial board game tree to be stored in computer memory. As a result, tree search algorithms based on partial board game trees work very fast. This approach can be applied only for specific games. Examples of such games are Atari-Go, Hex, and Gomoku. Experimental results relating to the solving of the Atari-Go game are presented here.