Seventy-eighth KGS Computer Go Tournament

Sunday December 4th, 2011

These results also appear on an official KGS page.


format20-round Swiss
board size9×9
time9 minutes plus 10/30s


The first round started at 16:00 UTC.

Result table

stv gomor MyGoF ManyF AyaMC pachi oakfo Orego
W13R B18R W112R B118R W1514 B1111 W117R BJ20 B11R W19R B115R W12T B110R W116R B14R W113R B119R W16R B114R W17R 19½195183½Winner
2gomorra4 B03R W08R B012R W018R
W1419 B014R W019R W17R B113R W116R B15R W09R B020R W11R B110R W117R W12R B111R B16R W115R 1221593
3MyGoFriend B0514 W0111 B017R WJ20 B0419 W114R B119R
W13R B064 W118R B076 W113R B0922 W115R WJ1 B18R WJ12 W12R B110R W116R 11½205½74½
4ManyFaces1 W01R B09R W015R B07R W013R B016R B03R W164 B018R
B14R WJ8 B114R W019R B12R W112R W110R B117R B15R W111R B120R 10½20466
5AyaMC B02T W010R B016R W05R B19R W120R W176 B013R W04R BJ8 W014R B119R
B16R W111R B018R W13R B115R W11R B112R W017R 10½200½78
6pachi W04R B013R W019R B01R W010R B017R W1922 B015R W02R B012R W06R B011R W118R
B15R W07R B116R WJ20 W13R B18R W114R 20240½
7oakfoam B06R W014R B02R W011R BJ1 W08R BJ12 B010R W017R B03R W015R W05R B17R W016R BJ20
W14R B19R W113R BJ18 W119R 717713½
8Orego12 B07R W06R B015R B02R W010R B016R W05R B011R W020R B01R W012R B117R B03R W08R B014R B04R W09R B013R WJ18 B019R

In the table above,
   0 is a loss
   1 is a win
   J is jigo
   left superscript is the player's colour
   right superscript is the round in which the game was played
   a subscript shows how the result was determined:
      R for resignation
      T for time
      F for forfeit
      a number for the points difference after counting.
All the 0s, 1s and Js are links to the game record.

Eight players entered.


At the start of round 1, pachi was present on KGS, but was playing a game with a human. As soon as this game was over, it joined its game with gomorra4 (which it lost). This happened because its operator was not at home when the tournament started, and had not configured it for tournament play. But he got home during round 1, and changed its configuration to tournament play.

In round 2, AyaMC had some problem in its game with stv. It restarted several times, but only ever played one move, and eventually lost the game on time.

In round 4, gomorra4 and MyGoFriend both passed, in a position that was won for gomorra4, with a dead group belonging to MyGoFriend still on the board. They disagreed on the status of the dead group, and the game entered the clean-up phase. Gomorra4 passed, allowing the dead group to be counted as alive. Thus gomorra4 is shown as winning this game by a smaller margin than it deserved.

stv vs MyGoFriend
"Clean-up" phase

In round 5, stv and MyGoFriend both passed in the position shown to the right. They disagreed on the status of the dead groups, and the game entered the clean-up phase. Stv played at A capturing four stones, and then they both passed, allowing the large black group and the single white stone on the right edge of the board both to be counted as alive. So this game is shown with stv winning by a smaller margin than it deserved. I assume that this happened because stv knew that the game would be scored in its favour after it had captured the four stones, and MyGoFriend does not support the clean-up phase at all.

In round 8, ManyFaces1 as white got a jigo with AyaMC. ManyFaces1 does not understand integer komi, and was playing as if the komi were 7½, so it would have thought the line it played was leading to a half-point win. It is possible that it could have "won" by more if it had understood the komi.

In round 12, MyGoFriend as white got a jigo with oakfoam. MyGoFriend also does not understand integer komi, and may have made a similar mistake.

In round 20, stv got a jigo with MyGoFriend. It had won all 19 of its previous games.

Thoughts on Integer Komi

During the tournament, I discussed integer komi, and the reasons why some programmers dislike it. I assume they don't like it because their bots don't support it, and they don't want to make changes that will only matter for the occasional 9×9 events I run on KGS.

Someone suggested that the problem of getting a bot to understand integer komi is generally not with the komi itself, but with the concept that a game can be drawn, as well as won or lost. I am sceptical about this, and would be interested to hear the views of programmers who have dealt with it. My thoughts are that, with an MC-based program that understands that the komi is 7, you don't need to have it understand that a game might be jigo; if a roll-out ends with a score difference of 0, you can just generate a random bit and use that to decide whether the game was won or lost. Then it will prefer jigoes to losses, and prefer wins to jigoes.

Annual points

Players receive points for the 2011 Annual KGS Bot Championship as follows:

Many Faces of Go2

Details of processor numbers, power, etc.

Aya, running on 6 cores of an i980X, at 3.3GHz.
Gomorra, running on an Infiniband cluster using 8×12cores running at 2.67 Ghz.
Many Faces of Go, running on the core i7.
MyGoFriend, running on a 20-core system: two i7s each with 6 cores and two i7s each with 4 cores, all about 3.3 mhz.
Oakfoam, running on a quad-core Intel Xeon E5520 with hyperthreading (so 8 threads).
Orego, probably running on one of the five nodes of a custom Linux cluster built by PSSC Labs: the node has two AMD Six Core Dual Opteron 2427 2.2 GHz (12 cores total), 8 GB RAM, Centos Linux.
pachi, running on an Intel Core2 Duo CPU E7200, two cores at 2.53GHz.
Steenvreter, running 46 threads each at 2.2 GHz, on a system whose use was generously provided by the Maastricht games and AI group.