Fifty-eighth KGS Computer Go Tournament

Sunday April 4th 2010

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


format15-round Swiss
board size9×9
time9 minutes plus 25/30s


The first round started at 08:00 UTC.

Result table


These results do not fairly represent the
strengths of CzechBot, ManyFaces1, and
pachi2, which did not play in all the rounds.

We welcomed a new competitor to this event: Fabian Linzberger, with his program 'kurt', playing on KGS sas 'kurtBot'. Named after Kurt Gödel, it is written in haskell and uses some heuristics followed by UCT tree search.


As there was an even number (16) of entrants, I left IdiotBot in the tournament.

In round 1, CzechBot showed evidence of a problem in its game against CrazyStone. It repeatedly connected, but never made a move, disappearing again within a second. This behaviour was to continue until round 11.

In round 2, SimpleBot beat PNUGo SGF. SimpleBot was once a fairly basic weak program, but its author Aloril has been improving it. PNUGo is a build of GNU Go.
       Break9 obtained a won game against WeakBot50k SGF, but lost it in the clean-up phase. Each player had one dead stone on the board, the players disagreed about the status of one of these stones (unfortunately the game record does not show which), and in the cleanup, WeakBot50k captured break9's stone, while break9 left WeakBot50k's on the board and passed, losing by half a point as a result.

In round 4, pachi2 connected to its game with kurtBot, but failed to make a move and timed out.
       Valkyria and Zen9 played an interesting game SGF, which valkyria won. I am not competent to judge the play, but none of the moves appeared to me obviously poor. This was one of Zen9's only two losses: the other, also to valkyria, was in round 15.

In round 5, pachi2 was able to connect to its game, but was still broken. However as it was playing against the also still-broken CzechBot this did not matter, it won on time without needing to make a move.
       LibegoBot and break9 failed to count their game SGF correctly, both accepting a dead white stone as alive. This did not affect the result, LibegoBot winning anyway.

In round 6, pachi2 was finally mended, and joined its game with LibegoBot late, with only 94 seconds of its nine minutes left SGF. It still managed to win.
       ManyFaces1 stopped playing halfway through its game with EricaBot, and lost on time. Its American operator was by now asleep, so it was not fixed, and timed out of all its remaining games.

Round 10 included an interesting and confusing (at least for me) game between EricaBot and Zen9 SGF. Zen9 won.
       CzechBot's operator, Petr Baudiš, connected to KGS, realised that CzechBot was not running properly, and started to fix it.

In round 11 CzechBot started playing.
       In its game with CrazyStone, valkyria used almost five minutes for moves 3 and 5 SGF, and later got low on time, started to play badly, and resigned.

EricaBot vs valkyria
Move 82.

In its round 14 game with valkyria SGF, EricaBot played move 82 as shown to the right. This is clearly a bad move, it achieves nothing, whereas capturing the single black stone at the top would be worth 24 points. I assume it did this because it realised that it could do nothing to win the game. If I am right, it shows that it had assessed the status of all the stones in the lower half of the board correctly (the black group is alive even though it does not have two eyes in the most obvious sense, and the white group is dead by "bent four in the corner"), so EricaBot's bad move actually showed that it had analysed quite a difficult situation correctly.

Round 15 included another interesting and complicated game between Zen9 and valkyria SGF. Again valkyria won.
       Although Zen9 lost two games, it was the clear winner of the tournament, with 13 wins from 15 games, two wins clear of its nearest rivals CrazyStone and valkyria. Having looked at its games, I feel this was thoroughly deserved, the two it lost were too difficult for me to understand.

Processor numbers, power, etc.

Aya, running on Xeon 2.6GHz 8cores
break, probably running on a single processor Intel(R) Celeron(R), 1.7Ghz
Crazy Stone, running on 4x2.2 GHz, AMD Opteron
MoGo, unstated but probably running on double-core AMD Athlon(tm) 64 X2 Dual Core Processor 4800+ (2.5GHz).
Erica, running on 8 cores, Xeon 2.26GHz.
running on Linux, 4GB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+
kurt, running on one core of an Intel i7 quadcore @ 2.67GHz
Libego, running on one core
Many Faces of Go, running on 16 cores, four 4-core Q8200 2.3 GHz.
pachi, running on 190 cores. It had been intended to run it on 1,500 cores, but this was reduced to 190 becasue of network issues.
Fuego r1096, running with four threads on an i7 920.
GNU Go, unspecified platform
running on one processor of a 4GiB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+, shared with WeakBot50k
valkyria, running on one processor of a i7-860 2.8 GHz
running on one processor of a 4GiB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+, shared with SimpleBot
Zen, platform not stated but possibly a Mac Pro 8 core, Xeon 2.26GHz.