電腦麻將程式TaKe的設計與實作

Abstract

近年在科技不斷進步之下，人工智慧電腦對局也不斷有新的發展與成果，越來越多新的技術被研發出來，舊有的技術也越來越成熟，但是相對於明確資訊的對局遊戲，不明確且帶有機率性的對局遊戲程式一直以來在學術上討論、探討的相關演算法較為少見，因此在這篇論文之中將會討論關於電腦麻將程式的人工智慧開發，所運用到的技術。 在本篇論文中，將會說明電腦麻將程式TaKe中所使用到的貝氏信賴網路演算法，以及人類麻將高手會使用的打牌技巧，如何來輔助電腦程式，找尋在過去麻將論文中較少提到的避免放槍機制，提供新的捨牌策略，以做到降低放槍機率。 目前該程式曾在TAAI 2013電腦對局比賽獲得銅牌，在TCGA 2014電腦對局比賽雖然獲獎未果，但是在比賽過程中該程式放槍次數是最低的，以玆證明使用的方法確實能達到期望中避免放槍的效果。在台灣十六張麻將雖然多以胡牌為最主要目的，但是在日本麻將因計分方式不同，因此未來若是有興趣開發日本麻將，或是在其他帶有機率性、不明確資訊遊戲的電腦程式開發，期望本論文能給予開發者一些啟發。

Because of the advances in science and technology, many computer games researchers continue to advance new methods and achievements in recent years. The existing technology has become increasingly mature. But relative to the perfect information games, programs that play imperfect information games have never been easier to compete with human players. And there is less paper talking about the algorithms that related to imperfect information games. In this thesis, we will discuss the development of computer AI program for playing mahjong. This thesis will explain most of the algorithms which have been used in our mahjong program “TaKe”. These algorithms include the skills that many master mahjong players have, and a Bayesian network algorithm to devise new strategies to change the original card throwing decision, and to reduce the winning rate of other players. Our program “TaKe” had won the bronze medal at TAAI 2013 computer game competitions. Though it hadn't won any prize at TCGA 2014 computer game competitions, but the number of the cards it was throwing result in a win to other players was the smallest, so that these methods are able to achieve the desired effect. We expect that the methods presented in this thesis can not only be used for the computer mahjong program, but alsofor any other imperfect information games with probability in the future.