Up Download

AlphaGo

AlphaGo is a computer go program developed by the Google company DeepMind. It was the first program to reach pro level. In Jan 2016 it was reported that AlphaGo had played a match against the European champion Fan Hui (in Oct 2015) and won 5-0. Simultaneously, a description of the used algorithms was published in the journal Nature.
David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lantot, Sander Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy Lillicrap, Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, Demis Hassabis, Mastering the game of Go with deep neural networks and tree search, Nature 529 (28 Jan 2016) 484–489.
In Mar 2016 it played a match against Lee Sedol and won 4-1.

Around New Year 2017 an improved version played 60 fast games on the Tygem and Foxwq servers against various opponents and won 60-0.

In May 2017 it played a match against Ke Jie and won 3-0.

Match against Fan Hui

date black white result #mv sgf
2015-10-05 Fan Hui AlphaGo W+2.5 272 sgf
2015-10-06 AlphaGo Fan Hui B+R 183 sgf
2015-10-07 Fan Hui AlphaGo W+R 166 sgf
2015-10-08 AlphaGo Fan Hui B+R 165 sgf
2015-10-09 Fan Hui AlphaGo W+R 214 sgf

Match against Lee Sedol

The game records under sgfc are commented by Fan Hui.

date black white result #mv sgf sgfc
2016-03-09 Lee Sedol AlphaGo W+R 186 sgf sgf
2016-03-10 AlphaGo Lee Sedol B+R 211 sgf sgf
2016-03-12 Lee Sedol AlphaGo W+R 176 sgf sgf
2016-03-13 AlphaGo Lee Sedol W+R 180 sgf sgf
2016-03-15 Lee Sedol AlphaGo W+R 280 sgf sgf

3 commented selfplay games

Three self-play games by AlphaGo, commented by Gu Li, Zhou Ruiyang, and Fan Hui.

date black white result #mv sgf
2016-02-29 AlphaGo AlphaGo W+2.5 274 sgf
2016-02-29 AlphaGo AlphaGo W+R 172 sgf
2016-02-29 AlphaGo AlphaGo W+R 202 sgf

New Year 2017

Below the games played by AlphaGo (under the name Magist or Master) on the Tygem server 2016-12-29,30,31 and on the Foxwq server 2017-01-02,03,04. The server handles of the opponents are known, but it is not always known which real life person uses that handle. Some names in the below are conjectured.

# date black white result #mv sgf
1 2016-12-29 19:01:34 Pan Tingyu Magist W+R 146 sgf
2 2016-12-29 19:20:57 Zhang Ziliang Magist W+R 174 sgf
3 2016-12-29 19:49:39 Magist Ding Shixiong B+R 151 sgf
4 2016-12-29 20:14:20 Xie Erhao Magist W+R 222 sgf
5 2016-12-29 20:51:02 Magist Yu Zhiying B+R 113 sgf
6 2016-12-29 22:02:39 Magist Li Xiangyu B+R 131 sgf
7 2016-12-29 22:40:09 Magist Qiao Zhijian B+R 163 sgf
8 2016-12-29 23:07:52 Han Yizhou Magist W+R 104 sgf
9 2016-12-29 23:30:11 Magist Meng Tailing B+4.5 275 sgf
10 2016-12-30 00:08:22 Meng Tailing Magist W+R 148 sgf
11 2016-12-30 10:02:17 Chen Hao Master W+R 170 sgf
12 2016-12-30 11:45:39 Wang Haoyang Master W+R 136 sgf
13 2016-12-30 12:06:51 Liu Yuhang Master W+R 144 sgf
14 2016-12-30 12:35:24 Master Yan Zaiming B+R 129 sgf
15 2016-12-30 13:02:15 Park Junghwan Master W+T 150 sgf
16 2016-12-30 13:52:26 Lian Xiao Master W+R 122 sgf
17 2016-12-30 14:45:18 Lian Xiao Master W+R 164 sgf
18 2016-12-30 16:15:35 Master Ke Jie B+5.5 228 sgf
19 2016-12-30 16:48:03 Ke Jie Master W+R 128 sgf
20 2016-12-30 17:14:02 Master Park Junghwan B+5.5 255 sgf
21 2016-12-31 09:30:30 Chen Yaoye Master W+5.5 270 sgf
22 2016-12-31 10:18:13 Chen Yaoye Master W+4.5 277 sgf
23 2016-12-31 13:09:13 Master Kim Junghyun B+R 135 sgf
24 2016-12-31 13:30:44 Master Park Junghwan B+R 223 sgf
25 2016-12-31 14:09:09 Park Junghwan Master W+0.5 261 sgf
26 2016-12-31 16:39:33 Master Yun Chanhee B+R 217 sgf
27 2016-12-31 16:50:18 Master Fang Tingyu B+R 215 sgf
28 2016-12-31 19:39:54 Master Meng Tailing B+R 163 sgf
29 2016-12-31 20:05:34 Mi Yuting Master W+0.5 311 sgf
30 2016-12-31 21:18:19 Tang Weixing Master W+R 186 sgf
31 2017-01-01 23:23:09 Master Li Qincheng B+R 179 sgf
32 2017-01-02 10:01:26 Gu Li Master W+R 154 sgf
33 2017-01-02 10:34:38 Master Gu Li B+R 191 sgf
34 2017-01-02 12:17:39 Master Dang Yifei B+R 149 sgf
35 2017-01-02 13:17:45 Jiang Weijie Master W+1.5 280 sgf
36 2017-01-02 14:41:33 Master Gu Zihao B+R 209 sgf
37 2017-01-02 15:16:54 Master Park Yeonghun B+R 173 sgf
38 2017-01-02 16:51:34 Master Tuo Jiaxi B+R 239 sgf
39 2017-01-02 19:40:22 Master Iyama Yuta B+R 135 sgf
40 2017-01-02 20:52:52 Meng Tailing Master W+2.5 274 sgf
41 2017-01-02 21:42:12 Kim Jiseok Master W+R 170 sgf
42 2017-01-03 10:22:04 Master Yang Dingxin B+R 125 sgf
43 2017-01-03 11:04:00 Master Kang Dongyun B+R 165 sgf
44 2017-01-03 13:51:15 An Sungjoon Master W+2.5 260 sgf
45 2017-01-03 14:46:57 Master Shi Yue B+R 167 sgf
46 2017-01-03 15:29:19 Lian Xiao Master W+R 144 sgf
47 2017-01-03 16:45:23 Master Tan Xiao B+R 191 sgf
48 2017-01-03 20:17:29 Park Junghwan Master W+1.5 270 sgf
49 2017-01-03 21:31:06 Weon Seongjin Master W+R 222 sgf
50 2017-01-03 22:13:23 Ke Jie Master W+R 178 sgf
51 2017-01-04 09:35:45 Zhou Junxun Master W+R 118 sgf
52 2017-01-04 10:40:25 Fan Tingyu Master W+R 202 sgf
53 2017-01-04 11:27:35 Master Huang Yunsong B+R 133 sgf
54 2017-01-04 15:02:36 Master Nie Weiping B+7.5 254 sgf
55 2017-01-04 16:05:42 Chen Yaoye Master W+1.5 267 sgf
56 2017-01-04 17:06:57 Master Cho Hanseung B+R 171 sgf
57 2017-01-04 19:35:45 Master Shin Jinseo B+R 139 sgf
58 2017-01-04 20:23:36 Chang Hao Master W+R 178 sgf
59 2017-01-04 21:18:23 Master Zhou Ruiyang B+R 161 sgf
60 2017-01-04 22:33:58 Gu Li Master W+2.5 235 sgf

The Future of Go Summit

On May 23-27, 2017 the "Future of Go Summit" was held in Wuzhen, China. It featured a 3-game match between AlphaGo and Ke Jie, a pairgo match between Gu Li + AlphaGo against Lian Xiao + AlphaGo, and a teamgo match between a team consisting of Chen Yaoye, Zhou Ruiyang, Mi Yuting, Shi Yue, and Tang Weixing against AlphaGo.

Match with Ke Jie

AlphaGo won 3-0.

date black white result #mv sgf
2017-05-23 Ke Jie AlphaGo W+0.5 289 sgf
2017-05-25 AlphaGo Ke Jie B+R 155 sgf
2017-05-27 AlphaGo Ke Jie B+R 209 sgf

Pairgo

The pairgo game was won by Lian Xiao + AlphaGo. The AlphaGo part of the losing team suggested to resign, but Gu Li refused at first, and agreed only when AlphaGo made the same suggestion a second time, some moves later.

date black white result #mv sgf
2017-05-26 Gu Li & AlphaGo Lian Xiao & AlphaGo W+R 220 sgf

Teamgo

AlphaGo won.

date black white result #mv sgf
2017-05-26 Chen Yaoye, Zhou Ruiyang, Mi Yuting, Shi Yue, Tang Weixing AlphaGo W+R 254 sgf

At the end the pro team played a tricky move in order to perhaps gain one or two points. AlphaGo generously gave them much more - it was so far ahead that that made no difference. The reaction:

facepalm

50 selfplay games

DeepMind published 50 more self-play games by AlphaGo, the first 10 on 2017-05-27, the other 40 on 2017-05-28.

# black white result #mv sgf
1 Alphago AlphaGo W+R 256 sgf
2 AlphaGo AlphaGo W+R 312 sgf
3 AlphaGo AlphaGo W+R 206 sgf
4 AlphaGo AlphaGo W+R 244 sgf
5 AlphaGo AlphaGo B+R 307 sgf
6 AlphaGo AlphaGo W+R 240 sgf
7 AlphaGo AlphaGo W+R 272 sgf
8 AlphaGo AlphaGo W+R 236 sgf
9 AlphaGo AlphaGo B+R 255 sgf
10 AlphaGo AlphaGo W+R 266 sgf
11 AlphaGo AlphaGo B+R 279 sgf
12 AlphaGo AlphaGo W+R 180 sgf
13 AlphaGo AlphaGo W+R 182 sgf
14 AlphaGo AlphaGo W+R 308 sgf
15 AlphaGo AlphaGo B+R 285 sgf
16 AlphaGo AlphaGo W+R 224 sgf
17 AlphaGo AlphaGo W+R 256 sgf
18 AlphaGo AlphaGo W+R 250 sgf
19 AlphaGo AlphaGo W+R 248 sgf
20 AlphaGo AlphaGo B+R 299 sgf
21 AlphaGo AlphaGo W+R 318 sgf
22 AlphaGo AlphaGo B+R 315 sgf
23 AlphaGo AlphaGo W+R 296 sgf
24 AlphaGo AlphaGo W+R 264 sgf
25 AlphaGo AlphaGo W+R 206 sgf
26 AlphaGo AlphaGo W+R 204 sgf
27 AlphaGo AlphaGo W+R 290 sgf
28 AlphaGo AlphaGo B+R 329 sgf
29 AlphaGo AlphaGo W+R 202 sgf
30 AlphaGo AlphaGo W+R 288 sgf
31 AlphaGo AlphaGo W+R 312 sgf
32 AlphaGo AlphaGo W+R 186 sgf
33 AlphaGo AlphaGo W+R 346 sgf
34 AlphaGo AlphaGo B+R 215 sgf
35 AlphaGo AlphaGo W+R 270 sgf
36 AlphaGo AlphaGo B+R 305 sgf
37 AlphaGo AlphaGo W+R 256 sgf
38 AlphaGo AlphaGo W+R 236 sgf
39 AlphaGo AlphaGo W+R 276 sgf
40 AlphaGo AlphaGo W+R 266 sgf
41 AlphaGo AlphaGo W+R 262 sgf
42 Alphago AlphaGo B+R 309 sgf
43 AlphaGo AlphaGo W+R 256 sgf
44 AlphaGo AlphaGo W+R 258 sgf
45 AlphaGo AlphaGo W+R 340 sgf
46 AlphaGo AlphaGo B+R 307 sgf
47 AlphaGo AlphaGo B+R 321 sgf
48 AlphaGo AlphaGo W+R 276 sgf
49 AlphaGo AlphaGo W+R 304 sgf
50 AlphaGo AlphaGo W+R 286 sgf

5 selfplay videos

date black white result #mv sgf video
2017-07-01 AlphaGo AlphaGo W+R 210 sgf video
2017-07-01 AlphaGo AlphaGo B+1.5 303 sgf video
2017-07-01 AlphaGo AlphaGo W+R 222 sgf video
2017-07-01 AlphaGo AlphaGo B+R 253 sgf video
2017-07-01 AlphaGo AlphaGo W+0.5 313 sgf video

AlphaGo Zero

In Oct 2017, DeepMind published (again in Nature) a description of AlphaGo Zero, a version of AlphaGo that was not trained on human games but had learned using self-play only.
David Silver, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, Yutian Chen, Timothy Lillicrap, Fan Hui, Laurent Sifre, George van den Driessche, Thore Graepel & Demis Hassabis, Mastering the game of Go without human knowledge, Nature 550 (19 October 2017) 354–359.
After 3 days of self-play (and 4.9 million games) it defeated the version that earlier had defeated Lee Sedol 100-0. This was the "20-block" version. Another version (with 40 residual blocks) also started from scratch, and after 40 days of self-play (and 29 million games) it defeated AlphaGo Master 89-11.

This publication comes with 83 SGFs, given below.

AlphaGo Zero (20 blocks) vs AlphaGo Lee

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Diagrams for the first 100 moves.

AlphaGo Zero scored 20-0. (And 100-0 over 100 games.)

AlphaGo Zero (20 blocks) self-play games

The 3-day training run was subdivided into 20 periods. The best player from each period (as selected by the evaluator) played a single game against itself, with 2h time controls. This produced the below 20 games.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Diagrams for the first 100 moves.

B won 8 times, W 12 times.

AlphaGo Zero (40 blocks) self-play games

The 40-day training run was subdivided into 20 periods. The best player from each period (as selected by the evaluator) played a single game against itself, with 2h time controls. This produced the below 20 games.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Diagrams for the first 100 moves.

B won 6 times, W 14 times.

AlphaGo Zero (40 blocks, 40 days) vs AlphaGo Master

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Diagrams for the first 100 moves.

AlphaGo Zero scored 17-3. (And 89-11 over 100 games.)

AlphaGo Zero Timeline

Three self-play games, played after 3h, 19h and 70h of training.

1 2 3

Diagrams for the first 80 moves (and further details).