Twenty-fourth KGS Computer Go Tournament

Sunday March 4th, 2007

Rules

Format

Formal division: six-round Swiss.
Open division: six-round Swiss.

Times

The first round started at 17:00 UCT for the Formal and 17:10 for the Open division.

Results

As usual, the tournament was held in two divisions, Formal and Open, with more restrictive entry conditions for the Formal division.

IdiotBot also registered for the Open division, and agreed to withdraw if this would make the numbers even.

The "real" names of the bots listed above, and of their programmers, are listed here: programs which have registered for KGS Computer Go Tournaments.

General

A newcomer to the event was Alan Blair with his program "thcGo". It played in both divisions, using the name "rcnn" in the Open.

Mango intended playing in the Formal division under its own name, and in the Open as MangoExp. Their registration email was sent at 16:45, but only reached me at 17:26 (as far as I can tell, the time was spent going from the Netherlands to the US on its way to my domain controller). Meanwhile, after the Formal division had started, its operator chatted to me in the Computer Go Room, and I was able to enter Mango for the Open division.

The MonteGNU in the Open division was closer to straight GNU code than the MonteGNU that played in the February tournament: Gunnar Farnebäck wrote "this time MonteGNU is only very marginally Monte Carlo, however, since that code is not quite up to playing fast on 19x19."

Last September Kocsis and Csaba Szepesvári published Bandit based Monte-Carlo Planning. The algorithm described there was implemented in MoGo. There was some doubt about its effectiveness for large boards and for fast games, but it was clear that it worked well for small boards particularly with slow time limits: MoGo has won nine of the eleven KGS bot tournaments held since September. However, this March event used full-sized boards and fast time limits. Some people expected MoGo to do less well, perhaps losing to more conventional programs such as ManyFaces, GNU Go, and Aya. They were wrong. MoGo won all twelve of its games.

Formal division

In round 1, CrazyStone beat AyaBot by 1½ points. MoGoBot and Orego had a disagremeent about the score, and applied the clean-up procedure correctly.

In round 2 MoGoBot beat GNU by 3½ points. ThcGo killed all of Orego's stones, but when they counted, Orego claimed that its two one-liberty groups were alive, and in the clean-up thcGo failed to capture them, thus winning by only 343½ points.

In round 3 MoGoBot beat ManyFaces1 by 3½ points. CrazyStone played out a losing ladder against GNU, and eventually lost all its stones.

In round 4 MoGoBot beat AyaBot by 2½ points. ManyFaces1 was well ahead against thcGo when, on move 156, it hit a time management bug and started passing. It started playing stones again with move 208, but by then thcGo had killed all its groups, and it was not able to save any of them. At the end of the game, with 355 black and no white stones on the board, the programs were unable to agree to end the game, and I assigned it as a win to Black.

In round 5 ManyFaces1 did not join its game with Orego. I kicked it in the hope of waking it up: after the standard five minute delay it appeared again, but still did not join its game. It eventually lost on time.

In round 6 ManyFaces1 again did not join its game. MoGoBot achieved a poisition in which it was about 100 points ahead of firstgo, but appeared in some danger of losing on time; rather than risk this, firstgo's owner Edward de Grijs resigned on its behalf. AyaBot was less fortunate: it killed all Orego's stones, but Orego coninued making pointless moves until AyaBot lost on time.

Open division

In round 1 slugGo achieved a won position against HouseBot. But at the game end SlugGo failed to enter clean-up mode correctly, allowing HouseBot to win.

In round 2 slugGo was unfortunate again, beating WeakBot50k by over 150 points and then losing on time.

Nothing remarkable happened in later rounds. MpGoBot19 won all its games.