Thirty-sixth KGS Computer Go Tournament

Sunday March 2nd 2008

These results also appear on official KGS pages: Formal Division, Open Division which link to the game records.

Rules

 Formal divisionOpen division
format5-round Swiss5-round Swiss
board size19×1919×19
rulesChineseChinese
komi
time43 minutes absolute43 minutes absolute

Times

The first round started at 16:00 UCT for the Formal and 16:10 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.

Formal Division   19×19

placenamewinsSOSSoDOS
1stCrazyStone51010
2ndAyaMC3122
3rdLeelaBot2130
4thGNU0150

Open Division   19×19

placenamewinsSOSSoDOS
1stStoneCrazy51212
2ndCzechBot3157
3rdMonteGNU3157
4thAyaMC23124
5thhcBot3113
6thSimpleBot371
7thWeakBot50k1120
8thbreak190150

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

We welcomed a new program, hcBot, by Michal Dvoracek. He explained that hcBot a very heavily modified version of GNU Go, and uses MoGo as an MC engine for the middle game.

As last month, no "official" MoGo entered, so I allowed CzechBot, Petr Baudiš's build of MoGo, to enter.

Formal division

The field of four players was small, but strong. Their KGS ratings (calculated from their 19×19 games, almost all against humans, on the KGS server) were shown as GNU 2d, CrazyStone 2k, AyaMC and LeelaBot 4k. These ratings are calculated from 0, 42, about 1500, and about 800 rated games, respectively. Thus the 2d is meaningless, but the other ratings are solidly based. KGS ratings are close to European (Czech-calculated) ratings, and somewhat stronger than AGA ratings.

In round 1, AyaMC and LeelaBot started fighting at the first opportunity. LeelaBot appeared to be trying to kill AyaBot's groups, but its ambition exceeded its execution, and AyaMC was able to win. GNU played a sensible-looking fuseki against CrazyStone, which preferred to make a central moyo. This moyo became solid territory, giving CrazyStone the game.

In round 2, AyaMC and CrazyStone began with messy fighting, and CrazyStone soon secured a lead. GNU again started with a sane fuseki, while LeelaBot played some inappropriate contact moves and many central stones. However LeelaBot was able to solidify its centre into territory, and win the game.

In round 3, the game between LeelaBot and CrazyStone did not look at all like a game between humans of similar strength – both players were much more interested in central stones than in territory, CrazyStone even more so than LeelaBot. Once the comprehensible fighting started, CrazyStone proved better at killing stones, and eventually won. The game between GNU and AyaMC was less remarkable. Aya was able to kill a group, and win the game.

In round 4, CrazyStone amd AyaMC engagaed in scrappy fighting in the middle of the board, while ignoring two of the corners. CrazyStone came out of this better, and won the game. GNU started with conventional territory-oriented moves while LeelaBot made a leaky central moyo. But GNU's corner territories also proved leaky, and it lost enough of them that LeelaBot was able to win by a few points.

CrazyStone (white) vs. LeelaBot
In round 5, CrazyStone and LeelaBot once more ignored the corners and played for central influence. The game appeared to have plenty of chances for both sides, in the position shown, when LeelaBot played the marked stone. This move achieves almost nothing, and is a long way below Leela's usual standard. When a "classical" program makes a move like this, its author can try to look for a bug which caused it – but as far as I know, you can't do this with an MC program like Leela. Can such a move be attributed to a statistical freak? If Leela plays 1,000 times from this position, are 999 of its moves better than this? Anyway, Leela eventually resigned this game.

GNU once more showed more knowledge of fuseki than Aya, was given no opportunity to use any knowledge of joseki, and lost its way once the actual fighting started.

Open division

In round 1, CzechBot (a MoGo build) played very slowly agaisnt MonteGNU. After 22 moves, CzechBot had used 25 of its 45 minutes, while MonteGNU had used 14 seconds. CzechBot got into time trouble, and lost. Its operator found that it had been configured for a 9×9 game, which may have confused its time manager.

After round 1 I removed IdiotBot from the draw, to make the numbers even and avoid byes. It would have been better if I had done this before round 1.

After five rounds, StoneCrazy had won all its games, and five programs each had three wins out of five.