experimental CGOS with "Pass Incentive" rule

[Hiroshi Yamashita]

Hi,
I started an experimental CGOS with "Pass Incentive" rule.
http://203.138.189.63/cgospass/9x9/standings.html

For recent AI, White winrate is too high with Chinese rule, komi 7.5.
2020 World Artificial Intelligence Go Competition adopted this rule.
Maybe this rule will be used in tournaments for a while except UEC Cup.

http://203.138.189.63/cgospass/9x9/cros ... tV100.html
This is same as kata1_b18s698v100 with
hasButton = true
ignoreGTPAndForceKomi = 7

Aya807_PassIn_p10k runs with this rule.

Server(currently 9x9 only) is
IP : 203.138.189.63
Port: 6909

Pass Incentive rule detais.
http://www.yss-aya.com/cgos/passincentive.html

[Rémi Coulom]

Thanks for your message.

It seems that the UEC Cup will use CGOS this year. Will the rules change to CGOS rules, or are they still Japanese rules? I would prefer CGOS rules.

[Hiroshi Yamashita]

UEC Cup will use Japanese rule.

CGF Open was also Japanese rule on CGOS.
http://203.138.189.63/cgf2023/
This CGOS server has modified to start game manually.
https://github.com/zakki/cgos/tree/admin_mode
WGo viewer is changed to update realtime.
UEC Cup will also use this CGOS.

I checked KataGo's recent selfplay games.
It seems black winrate is too low(0.4168) with Chinese rule + komi 7.5.

Chinese rule + komi 7.5 is 0.4168

But these two looks even.

Chinese rule + hasButton + komi 7.0 is 0.4869
Japanese rule + komi 6.5 is 0.4886

So maybe there is no big motivation to change UEC Cup rule.

Chinese rule + hasButton + komi 7.0 is almost same as "Pass Incentive + komi 6.5".

Code: Select all

Komi    Games B wins,Draw,W wins, Black winrate
5.5
   Jap 129269,70558,   53,58658, 0.5460
   Chi  51897,27654,   17,24226, 0.5330
   C_B  31603,15932, 4274,11397, 0.5717
6.0
   Jap 156315,70250,21224,64841, 0.5173
   Chi  79817,43565,  352,35900, 0.5480
   C_B  45536,24951,   15,20570, 0.5481
6.5
   Jap 155809,76086,   70,79653, 0.4886
   Chi 104121,57690,   55,46376, 0.5543
   C_B  54600,24659, 7232,22709, 0.5179
7.0
   Jap 127919,49956,17274,60689, 0.4580
   Chi 111528,41867,24870,44791, 0.4869
   C_B  54175,26362,   30,27783, 0.4869
7.5
   Jap  87917,37873,   30,50014, 0.4310
   Chi  97935,40798,   46,57091, 0.4168
   C_B  43918,17193, 5813,20912, 0.4577
8.0
   Jap  52111,18012, 6952,27147, 0.4124
   Chi  71015,30011,  343,40661, 0.4250
   C_B  29981,12972,   13,16996, 0.4329
8.5
   Jap  29185,11946,   15,17224, 0.4096
   Chi  43638,19373,   32,24233, 0.4443
   C_B  17685, 6091, 2251, 9343, 0.4081

Jap: Japanese rule
Chi: Chinese rule
C_B: Chinese rule + hasButton

* kata1 sgf (from 20040000 to 25530000, 1900000 games)

CGF Open(in Japanese)
http://hp.vector.co.jp/authors/VA012620/

[Rémi Coulom]

Interesting, thanks.

I do not think that balancing the chances of winning is so important, anyway. Making it easy for many programs to participate is more important. I don't like having to implement new rules for one tournament. There are so many more interesting things to do.