UCT Enhancements in Chinese Checkers Using an Endgame Database

Abstract

AbstractThe UCT algorithm has gained popularity for use in AI for games, especially in board games. This paper assess the performance of UCT-based AIs and the effectiveness of augmenting them with a lookup table containing evaluations of games states in the game of Chinese Checkers. Our lookup table is only guaranteed to be correct during the endgame, but serves as an accurate heuristic throughout the game. Experiments show that using the lookup table only for its endgames is harmful, while using it for its heuristic values improves the quality of play. This work is performed on a board with 81 locations and 6 pieces, which is larger than previous work on lookup tables in Chinese Checkers. It is a precursor to using the 500 GB full-game single-agent data on the full-size board with 81 locations and 10 pieces.KeywordsEvaluation FunctionLookup TableForward MoveGame StateGame TreeThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.