Forty-ninth KGS Computer Go Tournament

Sunday July 19th 2009

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


format9-round Swiss
board size19×19
time19 minutes plus 25/20s


The first round started at 16:00 UTC.



There were eight entrants, with two programmers submitting two programs each. Hiroshi Yamashita submitted his pre-MC/UCT version of Aya, from 2006, playing as AyaBot, as well his current AyaMC, with the agreement that if I wanted to remove one program to make the numbers even, I should remove AyaBot. Jason House sumitted hb07 as well as his current version, housebot; these two programs use "completely different board, playout, and search implementations".


In its round 2 against Orego, SGF housebot crashed. After a few minutes it was relaunched and rejoined the game, to find that its main time had run out, and its first Canadian-overtime period, with the very severe requirement to play 25 moves in 20 seconds, had started. It immediately played on a 1-1 point inside its own territory, and managed to get through this and two further overtime periods without losing on time. It was however far behind on the board.

AyaMC vs ManyFaces1
after move 109.

In round 3, the two eventual winners played the first of their two games of the event SGF. When Black, ManyFaces1, had played move 109, they reached the position shown to the right.
       To me, a human 2-kyu, it is immediately obvious that the two triangled groups are involved in a semeai. I am sure that one of those groups will live, and will do so by killing the other. I can't tell you the status of these groups, that is, I don't know who ought to win the fight and whether it is ko. It is difficult. As things stand, the white group has four liberties, the black group has three liberties but is involved in two kos so it will be hard for White to fill its liberties. But winning the semeai is worth 70 points, and I do feel confident that winning that fight must be much bigger than anything else on the board. If I were in that game, my subsequent moves would all be in that semeai, until I was confident that it was settled.
       However both players played elsewhere for four moves, then one move was played in the semeai, then nine more elsewhere, then one more in the semeai, then 35 more elsewhere, and so on. It seemed that the players did not regard this semeai as particularly important. My best guess is that they both reasoned something like this: "if I play an effective move in the semeai I win it 60% of the time. If I play elsewhere I win the semeai 40% of the time. The best move elsewhere is slightly bigger". This being a consequence of roll-outs of a game with a difficult semeai involving kos being pretty random with regard to the result of the semeai.
       Eventually, AyaMC won the semeai, and the game.

In the round 4 game between Fuego and housebot SGF, by move 300 Fuego had achieved a totally won position with a minute still left on its clock. Both players, knowing that no move would do anything to change the result, then played random meaningless moves for another 97 moves, at which point Fuego lost on time. If instead it had tried to bring the game to an end, by sealing off the territories, filling the dame, and passing, rather than playing pointless moves inside its own and its opponent's teritories, it could easily have won.

In the round 6 ManyFaces and Fuego started with a central fight, from which Fuego came out well. SGF. But ManyFaces1 won several later fights, and the game.

In the round 7 AyaMC and ManyFaces1 played their second game SGF, even though AyaMC had not yet played hb07 and ManyFaces1 had not yet played Orego. This is surprising, as a player should not play the same opponent a second time until it has played every opponent once. I do not know why the draw program chose this pairing.
       Anyway, this time ManyFaces1 beat AyaMC.

Processor numbers, power, etc.

Old pre-MC version of Aya, running on Opteron 852 2.6GHz (4 cores)
Aya, running on Opteron 852 2.6GHz x4 (4 cores)
Fuego, unspecified but may have been running on an 8-core 2.66Ghz Xeon.
an old version of HouseBot, before a total rewrite, running on one core of an AMD Athlon(tm) 64 X2 Dual Core Processor 5000+
HouseBot, running on Intel(R) Core(TM)2 Duo CPU T5450 @ 1.66GHz
Many Faces of Go, running on a Core2 Quad Q8200 2.3 GHz
running on 2 x 3 GHz Dual-Core Intel Xeon
running on a single Pentium-M 1.4GHz processor