Fourth 2011 Slow KGS Computer Go Tournament

In the table above,
   0 indicates a loss
   1 indicates a win
   a superscript indicates the round in which a 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.

The numbers in this table may not add up as you expect. This is because the "Wins" column incudes byes.

Seven players entered. They did not include Zen, which also did not enter the October KGS bot tournament. Hideki Kato emailed me explaining that as its position as KGS Annual bot champion was already guaranteed, its human and computer resources would be used in other ways, allowing others to win the last three KGS bot tournaments of the year.

Pachi entered as 'pachi' instead of as 'pachi2', as it was running on a much smaller system than usual.

Results

In round 1, MyGoFriend, playing black against PNUGo, did not make a move for the first five minutes. Its operator disconnected it, and logged in using its account to check what was happening. He decided that it had not received a 'genmove' command from the server, did something to remedy this, and let it log in again. It started playing, and eventually beat PNUGo by 9½ points.

However, when the server detected a login by a human to the account of a competitor in a bot tournament, it decided that this was an attempt to cheat, and declared the game forfeit, and PNUGo the winner. The results table at the KGS page for the tournament shows PNUGo as winning this game, and calculates the scores, SOSs, and SODOSs accordingly.

I believe that this was not an attempt to cheat. Moreover, after the "forfeit", the game continued, and MyGoFriend won it. I discussed this with PNUGo's operator, and we agreed that the game should be counted as a win by MyGoFriend. The table above is based on this. I have calculated the scores, SOSs, and SODOSs above myself, I hope correctly.

Though the KGS tournament system believes that PNUGo won this game by forfeit, the KGS database of past games believes that MyGoFriend won it by 9½ points. I was surprised that two components of the KGS server can have inconsistent views about the result of a game, and at the time of the tournament I assumed it was due to a bug in KGS. I reported it as a bug, and discussed it with KGS's programmer, 'wms'. He assured me that this behaviour is intentional. I therefore describe it here at length, in case a similar incident happens again.

In future, if the KGS tournament system and the KGS database show different results for a game, I shall give priority to the tournament system. An operator who logs in to his bot's account during a game will receive a loss by forfeit. I cannot explain why MyGoFriend failed to receive a 'genmove' command; but other bots do not have this problem, so I conclude that the fault is with MyGoFriend, not with KGS.

I still reserve the right to overrule a decision made by the KGS tournament system, in cases where I believe it has made the decision in error.

In round 9, MyGoFriend (playing against ManyFaces1) and pachi (playing against GoKnot) both joined their games normally, but neither of them moved for 37 minutes. I discussed this with MyGoFriend's operator, and recommended that he should not log in to its account himself because of the forfeit problem experienced in round 1. We agreed that I should kick it from the server, in the hope of waking it up.

It reconnected a few minutes later, and started to play normally. Within another five minutes, pachi (whose operator was shown on KGS as present but inactive) also woke up and started to play normally.


gomorra4 vs ManyFaces1
Moves 17-26.


gomorra4 vs ManyFaces1
Moves 67-75.

In its round 12 game with gomorra4, ManyFaces1 as Black chased a white group in a ladder, initially not seeing that the ladder was broken by a white stone, as shown in the diagram to the left. Later in the game, it did the same again, as seen in the diagram to the right. But ManyFaces1 was still able to win the game, thereby winning the tournament with no loss.

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

Many Faces of Go	8
gomorra	5
pachi	3
MyGoFriend	2
GoKnot	1

Details of processor numbers, power, etc.

GoKnot: GoKnot, running on a Phenom II (6 core) at 3.3 GHz, not overclocked.
gomorra4: Gomorra, running on a cluster using 4x12 cores. Cores are running at 2.67 Ghz.
ManyFaces1: Many Faces of Go, running on an core2 quad.
MyGoFriend: MyGoFriend, running on unspecified hardware.
Orego12: Orego, running on one of the five nodes of a custom Linux cluster. The node has two AMD Six Core Dual Opteron 2427 2.2 GHz (12 cores total), 8 GB RAM, Centos Linux.
pachi: pachi, running on a 6-core AMD Phenom 1090T.
PNUGo: GNU Go, running on a 1-core system.