電腦麻將程式MahJongDaXia的設計與實作

吳俊緯

在人工智慧的電腦對局領域中，麻將是一個相當特別的項目，由於多玩家、不完全資訊及機率性三個要素使得其他項目的搜尋演算法在麻將中變得難以實現，也因為其困難與價值，使得國內外有許多優秀的人才投入這個領域。
在本論文所研發的程式MahJongDaXia，用純規則導向的架構與純機率性的審局方式達到所期望「正確的捨牌」。在文中將探討過去各電腦麻將程式的架構及應用技術，並詳述MahJongDaXia的程式結構，所採用的7個規則來做為決策函式的核心，說明這些規則對於特殊牌型的處理，關於開發過程對於使用的結構的改進原因及優劣。MahJongDaXia為一個從無到有的電腦麻將程式，開發實踐尚短，也在TAAI 2015、TCGA 2016與ICGA2016競爭中取得銀牌，未來會以程式完善為目標改進。

The Design and Implementation of Mahjong Program MahJongDaXia

by

Chun-Wei Wu

In the field of artificial intelligence in computer game programming, Mahjong is a very special item. Due to three factors: the multiplayer, incomplete information, and probabilistic outcomes, the conventional search algorithms used in other items become impractical in Mahjong. But also because of its difficulty and value, there are a lot of researchers promoting this area in recent years.
In this thesis, we introduce our program MahJongDaXia that uses the rule-oriented approach and the probability computation to achieve the desired "correct choice". We investigate the structures and technologies of previous MahJong programs and MahJongDaXia in detail. We describe the 7 rules that are used by MahJongDaXia to be the core of its decision functions, how to process the special card patterns by rules, and the pros and cons of these technologies used in different versions of our program. Although it does not have very long development time, it stood out from competitions in TAAI 2015, TCGA 2016, and ICGA 2016 and got the silver medals. We will be continuing to improve the program in the future.

[1] 林典餘，麻將人工智慧之研究。2008，國立交通大學研究所碩士論文。
[2] 沈庭瑋，電腦麻將程式TaKe的設計與實作，2014，國立臺灣師範大學資工所碩士論文。
[3] 陳新颺，電腦麻將程式ThousandWind的設計與實作，2013，國立臺灣師範大學資工所碩士論文。
[4] 郭桐甫，電腦台灣麻將程式設計與實作，2016，第六屆電腦對局研討會。
[5] 莊凱閔、陳玥汝，電腦麻將演算法及相關議題之研究，2007，第十二屆人工智慧與應用研討會。
[6] 莊立愷，唐士傑，麻將人工智慧之研究，2015，第二十屆人工智慧與應用研討會
[7] 曾海洋，蒙地卡羅麻將程式設計與實作，2015，第五屆電腦對局研討會。
[8] Wan Jing Loh, AI Mahjong, 2009. http://cs229.stanford.edu/proj2009/Loh.pdf