Forty-sixth KGS Computer Go Tournament

Sunday April 6th 2009

These results also appear on an official KGS page which links to the game records.


 Formal division
format6-round Swiss
board size19×19
time29 minutes plus 25/20s


The first round started at 08:00 UTC.


Formal Division   13×13


We welcomed a new entrant to these events: Yamato's 'Zen', playing as 'Zen19'. The details of Zen are secret, but Yamato (the pseudonym of an individual programer working in Japan) has revealed that it uses shape patterns generated by minorization-maximization (as Crazy Stone does), uses UCT, and does not use progressive widening. It contains a lot of hard-coded Go knowledge. It became known when it started playing on KGS on March 26th, and rapidly attained a 1-dan rating, running on a four-core system.

MoGo did not register, so Petr Baudiš registered his build of MoGo 'CzechBot' on the understanding that he would withdraw it if I received an official entry from MoGo.


Zen19 vs Fuego
Move 60.
In round 1 valkyria and StoneGrid each tried to claim half the board. StoneGrid's half proved slightly larger, and valkyria19 resigned.
       CzechBot and AyaMC played a fighting game, with each killing several small groups. Then AyaMc killed a larger group and CzechBot resigned.
       The game between Zen19 and Fuego SGF promised to be the most interesting in this round, as they were the two most highly rated on KGS, 1-dan and 2-kyu respectively. Zen19 started by showing a disdain for territory, but it later acquired plenty by killing black groups. The first of these black groups died with the sequence shown to the right. Now that the leading bot players are stronger than me, I am reluctant to comment on their moves; but I was impressed by move 60. Zen could instead have played at 61, to save its three triangled white stones (in a rather ugly sequence which would give black some useful strength). But it preferred the move 60 shown, which threatens to save those three stones by capturing two black stones in a ladder; and when black captured the three stones with 61, it followed up by playing 62, killing black's triangled group.

In round 2 Zen19, playing against valkyria19, started quite differently, making solid corner teritory. It won with little fighting. StoneGrid played out a long losing ladder agaisnt AyaMc, in an already lost position, and lost. CzechBot beat Fuego.

In Zen19's round 3 game against AyaMC, the first five moves looked normal, but Zen19 then played on a 7-8 point. This provoked a fight, which Aya won. Zen19 then used its outside stones to kill two of AyaMC's groups. AyaMC also had some successes, but had to resign when it lost another group.
       CzechBot and StoneGrid appeared to ignore each others' moves, CzechBot claiming rather more than half the board and StoneGrid the rest. But StoneGrid proved better at enlarging its half, and the game became close. The players failed to agree on the dead stones (though there was nothing at all difficult), and after a tedious clean-up phase, CzechBot won by 4½ points.
       Fuego killed a group in an otherwise balanced game against valkyria19, causing it to resign.

AyaMC vs valkyria19
Move 28.
In round 4 there was little early interaction between StoneGrid and Fuego, as each claimed half the board. Fuego's half proved larger, and StoneGrid resigned.
       Playnig against Zen19, CzechBot made a large moyo while Zen19 claimed all four corners. Zen19 then invaded the moyo, but this was not enough, CzechBot still had more territory.
       In the game betwen AyaMC and valkyria19 SGF, AyaMC played move 28 as shown in the figure to the left. This is, I believe, a mistake, and not the kind of mistake that a human 5-kyu would make. This move almost captures the two triangled black stones. Capturing them properly, by playing 28 one point further north, must be better. A few moves later valkyria19 misread a ladder and lost the triangled stones anyway; but still won the game.

In its round 5 game with Zen19, StoneGrid disappeared for a while, but reconnected and resumed play. later it chased Zen19's stones in a ladder that did not work, and lost. Fuego made a large white moyo in its game with AyaMC, large enough to win the game. CzechBot won a fight at the top of the board, killing valkyria19's group, and valkyria19 resigned.

After five rounds, each entrant had played each other once, so it would have been a good time to end the tournament. CzechBot would have ben the winner, equal with Zen19 on wins and ahead on SoS, the first tiebreaker in KGS tournaments. However, the format of a KGS tournament has to be specified in advance, before the number of entrants is known, and I had specified six rounds. So one further round, round 6, was played. In this, the two leading players were paired for a second time. They played with the same colours (which seems to me to be a weakness on the pairing program, but not one serious enough to justify asking the programmer to fix). The game was the same as their round 4 game for the first eleven moves, with all the moves on the third and fourth lines. But in both games, CzechBot played move twelve near the centre, choosing a 9-9 point in round 4 and a 9-10 point in round 6, so the games then diverged. CzechBot won again, after a well-fought game.

Processor numbers, power, etc.

Aya, running on Xeon X5355 2.66GHz 2CPU (4 cores on 1CPU, so all 8 cores)
MoGo, running on double-core AMD Athlon(tm) 64 X2 Dual Core Processor 4800+ (2.5GHz).
was running on a 16-core AMD Opteron, 2.7GHz.
was running on an Intel Core 2 Duo L7700 1.8GHz.
valkyria, running on a single processor 3Ghz P4
Zen, running on a quad-core AMD Opteron