Building Connect6 opening by using the Monte Carlo tree search

Abstract

An opening book is an important part in most game-playing computer programs. The purpose of the research aims to construct an Opening-Book system of Connect6 game. In this study, it develops a real system that can apply Opening-Book system of Connect6 to Kavalan. However, the method in which constructed the system is not by profession's domain knowledge, and it is constructed by numerous end games. The study collects various end games to construct positions of opening book from public competitions, including self-games. The study incorporates the previous results into the design of the tree structure of Connect6 opening: the research of Bitboard design and bitwise computing of Connect6, revised algorithm of MCTS to fit the property of sudden-death, and the experience of software development of Connect6 game. It plays an important role of developing Connect6 opening to combine the building of Opening-Book system and the search algorithm of Connect6. In addition, it can show the overall efficiency only if the search algorithm and Connect6 opening are perfect match. Hence, two methods balance the advantage and disadvantage to achieve the greatest accomplishment. The study has finished the analysis of Connect6 board and the design of Connect6 opening. Besides, it also finished the development of the Opening-Book system of Connect6 game, and attached it to the MCTS of Connect6. With the increasing of positions saving in the Connect6 opening, Kavalan already greatly reducing the time spent on opening-game. Therefore, the results of the research greatly enhance the search efficiency of Kavalan.