暗棋審局函數的設計及實作

Abstract

電腦對局是一個被研究多年的領域，許多知名演算法（如 alpha-beta 剪枝與蒙地卡羅樹搜尋）均與此領域相關。 本論文所使用的暗棋程式 Yahari 採用 minimax 演算法結合 alpha-beta 剪枝進行搜尋。minimax 演算法的表現高度取決於審局函數的品質。目前，Yahari 的審局函數主要依據子力價值進行評估。在遊戲中期，由於子力交換頻繁，該方法能夠提供良好的判斷。然而，在遊戲後期，子力變動減少，僅考慮子力價值已不足以準確評估局勢，因此需要考慮額外因素。本論文將探討如何透過棋型結構來改良審局函數，以進一步提升程式的對局表現。

Computer games have been a well-studied field for many years. Many famous algorithms, such as alpha-beta pruning and Monte Carlo tree search, are closely related to this domain. The Chinese dark chess program used in this study, Yahari, employs the minimax algorithm combined with alpha-beta pruning for its search engine. The effectiveness of minimax heavily depends on the quality of an evaluation function. Currently, material value is the main factor considered in Yahari's evaluation function. This performs well in the middle game since piece exchanges occur frequently. However, in the endgame, as piece exchanges decrease, considering only material value is insufficient for an accurate evaluation of the position. Therefore, additional factors should be taken into consideration. This thesis explores how to improve the evaluation function by incorporating piece positional structure to further enhance the program's performance.