Fifty-second KGS Computer Go Tournament

Sunday October 4th 2009

These results also appear on an official KGS page which links to the records of all the games.


format8-round Swiss
board size19×19
time29 minutes plus 25/30s


The first round started at 08:00 UTC.

Result table


As last month, Aloril entered FuegoAl and GnuGoAl, unofficial versions of Fuego and GNU Go. He also entered his usual collection of weaker bots, and allowed me to remove IdiotBot if it would make the numbers even. Later an official version of Fuego was entered by Broderick Arneson, so I withdrew FuegoAl.

Petr Baudiš entered CzechBot, an unofficial version of MoGo, as well as his own pachi.


Orego vs WeakBot50k
Moves 367-369.

In round 1, ManyFaces2 joined its game against AyaMC eight minutes late, while its 32-core system was set up. It lost the game. SGF

Stv started very slowly against Fuego, but managed to speed up when it got into the overtime of 25 stones in 30 seconds. However it did not speed up quite enough, playing only 24 stones before the 30 seconds expired. SGF

In the game between Orego and WeakBot50k, the sequence shown to the right occurred. SGF Orego then tried to recapture 369, and was not allowed to, because of the positional superko rule. It then timed out. Orego's author Peter Drake reported

Here's what I think happened:

That move got a high heuristic weighting, probably because Orego had gone into the new "coup de grâce" mode that suggests playing near enemy stones without too many liberties. The move was never actually tried in the playouts. When it came time to choose a move, g3 happened to have the most wins. I was checking for superko before adding anything to the tree, but not verifying that the "best" move was actually legal.

That's fixed now. Sorry about the inconvenience.

Unfortunately, Orego never recovered from this, and was absent from the server for the rest of the tournament, losing all its games by default.

In round 2, wei2go started the game against ManyFaces2 by playing on a19; and resigned for move 5. SGF

Stv achieved a thoroughly won position against SimpleBot, but with little time left. However it managed to play its last 50 moves in less than a minute, securing its win. SGF

GnuGoAl killed all of WeakBot50k's stones, but appeared to have its time mis-set, so it was in danger of losing on time. However it also speeded up enough to secure its win. During the late stages of the game, when then only issue was whether it would be able to play fast enough, it did not make solid moves and capture dead stones as a human would have done, it created pointless kos as if trying to prolong the game. I guess there is little point in creating and testing code to do such things in a sane-looking way, as long as its time-handler is robust and the is no lag. SGF

AyaMC had its only loss of the tournament, against CzechBot. SGF

CzechBot vs Fuego
After move 9.
pachi vs stv
After move 9.

In round 3, CzechBot-Fuego (far left) SGF and pachi-stv (left) SGF both started with strange-looking high fusekis.

In round 4, wei2go played move 1 on a19 against GnuGoAl, and resigned for move 3. SGF

ManyFaces2 achieved a competent win against stv. SGF

After round 4, there were five players on three wins: AyaMC, CzechBot, Fuego, ManyFaces2 and GnuGoAl.

SimpleBot vs wei2go
Moves 185-186.

In round 5, wei2go once more started with a19, this time against SimpleBot. It seems it was encouraged by SimpleBot's direct response at a18, and it played a reasonable game until it reached the sequence shown to the right. Then wei2go tried to recapture, and because of the positional superko rule, it was not allowed to do so. Like Orego earlier in the tournament, it timed out. SGF

After four moves of its game against GnuGoAl, CzechBot went to sleep and stopped doing anything. I decided to try to wake it up by kicking it. This is a difficult decision for me - it disconnects the bot from the server, and it does not rejoin again for five minutes, so if the bot was in fact thinking, it is a disaster for it. However in this case, it turned out that something had gone wrong with CzechBot. It did reconnect after five minutes, but still played no moves, and lost when it timed out. SGF Its operator Petr Baudiš explained

CzechBot hung after "selfTrainNewMaster(...) --- selfTrainInitiation "
kicking it didn't help
"FINEST: Still an outstanding command", I will wait until the system is idle before making a new game.

AyaMC vs StoneGrid
Moves 161-164

In round 6, Fuego timed out in a totally won position against WeakBot50k. This was not a case of bad time management, Fuego had five minutes left and nothing significant to think about, it just stopped responding. SGF
       Its creator Martin Müller told me that it had also had a problem, like CzechBot above, with "FINEST: Still an outstanding command". I do not know what this means, it suggests to me that there is a command in the protocol which is still "outstanding", i.e. not yet implemented.

In the game between AyaMC and StoneGrid, they reached the position shown to the left. StoneGrid is probably losing, but it has killed a large white group at the bottom left, and may be able to win if it can establish a live group near the top edge of the board. However, it played the sequence shown, which leaves a ko to decide the fate of the formerly dead white group and maybe also of the formerly alive black group above it. Neither player fought this ko competently, but after Black had tried and failed to establish a group on the top edge, it resigned. SGF

In round 7, Fuego was drawn against the absent Orego. After waiting for more than 20 minutes, Fuego vanished. later, Orego timed out and lost.

Wei2go started well in its game against WeakBot50k, but on move 192 it resigned in a position in which it still had plenty of winning chances. SGF

After round 7, AyaMC and ManyFaces2 were each on six wins, CzechBot was on five.

In round 8 AyaMC was drawn against pachi, and won; while CzechBot beat ManyFaces2. So AyaMC won the tournament with seven wins, ahead of ManyFaces2 and CzechBot each with six.

Orego and Fuego were both still absent. So when wei2go started on a19 in its game against Fuego, it was enough for it to win on time.

Processor numbers, power, etc.

Aya, running on Opteron 852 2.6GHz x4 (4 cores)
MoGo, running on double-core AMD Athlon(tm) 64 X2 Dual Core Processor 4800+ (2.5GHz).
was running on an 8c Intel Xeon 2.5GHz with 8GB of memory
was GNU Go, with no MC code, running on Aloril's Linux-based laptop, 2GiB RAM, Intel Celeron M CPU 530 @ 1.73GHz
Many Faces of Go, running on a 32-core Xeon 3.2GHz: 4 nodes, 8 cores per node, connected by a 40Gbps network
running on 2 x 3 GHz Dual-Core Intel Xeon
runing on three cores of i7 920 @ 2.67GHz.
running on one processor of a 4GiB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+, shared with WeakBot50k
Steenvreter, running on an Intel Core2, Q6600
was probably running on an Intel Core 2 Duo L7700 1.8GHz.
running on one processor of a 4GiB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+, shared with SimpleBot
Wei 2 Go, running on MacBook 2.0 GHz dual core Intel, 1 Gb DDR2 RAM