wjx0912 / dentsu_engine_tutorial Goto Github PK

Monte-Carlo Go samples.

Hiroshi Yamashita

2015/11/14

2016/11/07 go10.c - go15.c, fixed "play b p3" makes "pass" in 19x19.

01/go01.c * print Go board

02/go02.c * teaching Go rule

03/go03.c * ramdom player

04/go04.c do playout by random player

05/go05.c * calculate win or loss in final position

06/go06.c primitive monte-carlo Go program

07/go07.c primitive monte-carlo Go program in NegaMax style

08/go08.c search by UCT

09/go09.c selfplay UCT and primitive MC

10/go10.c GTP. 3x3 pattern, Progressive Widening

11/go11.c RAVE

12/go12.c Zoblist Hashing

13/go13.c print playout owner

14/go14.c Criticality

15/go15.c Time limit play within 10 minutes

NOTE : * source includes err intentionally. Correct source is in "fixed" directly.

RAVE (go11.c) is strongest.

This samples are for UEC Computer Go Workshop at August 2015.

如果时间允许，这15课会写个教程（也许没必要，因为代码很容易理解）：主要讲了mcts,uct(ucb1)算法，简单模式识别，

RAVE优化，zobrist hash，对杀剧烈程度表示，棋子强弱表示（可以参考实现形势判断）。

不足之处是没有alpha-beta剪枝优化搜索树（论文中有描述），有兴趣的同学可以研究pachi源码

原版mcts/uct paper: http://www.cs.utexas.edu/~pstone/Courses/394Rspring13/resources/mcrave.pdf

QQ群：732665928，非技术勿聊