[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

Journal of Information Science and Engineering, Vol. 37 No. 2, pp. 425-439

Seed Optimization Framework on Draughts

1Department of Computer Science and Information Engineering
National Dong Hwa University
Hwalien, 974 Taiwan

2TAO, INRIA, France
E-mail: 810221001@gms.ndhu.edu.tw; fabien letouzey@hotmail.com;
olivier.teytaud@inria.fr; sjyen@mail.ndhu.edu.tw

Seed optimization has been successfully tested on many games such as Go, Domineering, Breakthrough, among others. Fixed seeds can outperform random seeds by selecting locally optimal seeds as different playing policies. In this article seed optimization has been tested for the Draughts program Scan. We provide a framework which can optimize a draughts program for competition. It does not affect the original program structure, so it improves the strength with no modifying algorithm and no penalty when executing. With the new Best Promise Seed framework, the win rate can be improved by replacing the random seeds with some pretested locally optimal seeds. The optimized program won the championship in the Computer Olympiad in 2015 and 2016. It shows that self learning methodology improves the strength of Scan against other competing programs. In addition, better locally optimal seed(s) may be discovered with a longer learning time, so further strength improvement is possible. All current draughts programs and other different game programs might gain benefit from this framework. 

Keywords: draughts, seed optimization, BestSeed, policy optimization, machine learning

  Retrieve PDF document (JISE_202102_10.pdf)