Fortieth KGS Computer Go Tournament

Sunday July 6th 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 UTC for the Formal and 16:05 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
1stCrazyStone51414
2ndAyaMC4127
3rdLeelaBot3156
4thHouseBot3113
5thGNU2131
ManyFaces12131
7thFirstGoBot1100
8thOrego0120

Open Division   19×19

placenamewinsSOSSoDOS
1stStoneCrazy51111
2ndManyFaces24138
3rdAyaMC23156
4thHBotSVN383
5thMonteGNU2153
6thSimpleBot2131
7thWeakBot50k1120
8thwei2go0130

We welcomed one new player, Basti Weidemyr's "Wei 2 Go". At present it uses a simple search algorithm – later versions are likely to incorporate MC techniques.

WeakBot50k was still shown on the server as having an undeserved rating of "2d?". However there was a genuinely strong set of entrants for this event – CrazyStone was 1k, and AyaMC and LeelaBot were 3k, all these ratings being earned by playing rated games against all comers on KGS.

The two versions of Many Faces of Go were different. ManyFaces1 was using MC/UCT methods, while ManyFaces2 was using alpha-beta search.

Formal division

In round 2, AyaMC and GNU played an interesting game SGF. By the end of the game, GNU had territory in all four corners and owned 64 of the 1-line points, while AyaMC owned only eight of them. But AyaMC with its central territory was ahead, by 2½ points.

Also in round 2, the game between HouseBot and ManyFaces1 was even more interesting SGF. HouseBot's fuseki was eccentric, and by move 160, ManyFaces1 had killed several groups and secured all four corners and most of the rest of the board, and was over 100 points ahead. But HouseBot slowly made gains, while ManyFaces1 did nothing to stop it, in the manner of a UCT program that is confident it has won. After move 276 ManyFaces1 was only half a point ahead; and then played two of its own stones into atari, giving HouseBot the lead. When both players finally passed, HouseBot was 2½ points ahead.
       When the players were asked by the server to submit their lists of dead stones, ManyFaces submitted a (correct) list, and HouseBot submitted no list (HouseBot never submits a list). What ought to happen when one player submits a list and the other does not, is that the server believes the list it does receive, and scores the game accordingly; and at one time, this was what did happen. Thus HouseBot's failure to submit a list is trusting (it leaves itself open to being swindled by an unscrupulous opponent) but legal.
       However, for some months now, the server has been behaving incorrectly. When only one player has submitted a list of dead stones, it acts as if the other player has claimed all its stones were alive (Jason House had reported this bug to me, but I had failed to read his email properly, and did not act on it. I have now reported it to wms, the programmer of KGS, and he has acknowledged that it should be fixed). This causes play to resume, until the status of all the "disputed" stones has been established. So in this game, play was resumed, but HouseBot, which had only 30 seconds left, timed out without moving, and thereby lost. David Fotland then generously asked me to treat the game as a win for HouseBot, which I did. The game is shown as "forfeited" by ManyFaces1, not the designation I would like, but I have no power to change it.

In round 3, FirstGoBot took the lead against HouseBot SGF, and by move 164 it was over 100 points ahead. However for move 165 it passed, and continued to pass whenever it felt that its lead was still big enough. It overdid this, and continued passing or making meaningless moves, while HouseBot enlarged its territory and killed FirstGoBot's group in the lower right. Soon after this, FirstGoBot resigned.

In round 5, CrazyStone played another effective central territory game SGF, against GNU. Early in the game it scattered stones in the middle of the board; and then it let GNU take all four corners and most of the sides, while securing enough central territory for itself to win by 1½ points.

Open division

I removed IdiotBot from the draw before the draw for round 1, to make the numbers even and avoid byes.

AyaMC2 vs HBotSVN
The position after move 277, marked.
In round 2, HBotSVN played the marked move against AyaMC2 SGF in the position shown to the right. This is a good move, as it converts the corner to seki, and so deprives AyaMC2 of territory there. It is worth nine points in gote (11 points, using Chinese counting). It is the biggest move on the board, but not enough to win the game. AyaMC2 could have prevented this seki with any of its previous few moves, but did not do so because it knew it was winning anyway.
       However, HBotSVN does not recognise seki in its playouts (it is very difficult to recognise sekis correctly and fast). It now thought that the upper right would all belong to whichever player "won" there by its opponent filling its own penultimate liberty first (HBotSVN's playouts do not consider passing when there are neutral points to fill). As it considered that each player was equally likely to do this, and knew that capturing in the top right was enough to win the game, it believed that by making the marked move it had increased its chance of winning to around 50%.
       However neither player ever filled its own penultimate liberty there. The seki was still on the board at the end of the game, and AyaMC2 won.

Also in round 2, wei2go did not respond to move 28, and its opponent WeakBot50k won when wei2go timed out.

In round 3, wei2go did not respond to move 204, and its opponent SimpleBot won when wei2go timed out.

In round 4, wei2go did not respond to move 219, and its opponent HBotSVN won when wei2go timed out.

In round 5, wei2go did not respond to move 27, and its opponent StoneCrazy won when wei2go timed out.

 

 

Processor numbers, power, etc.

AyaMC and AyaMC2
Aya, running on Xeon X5355 2.66GHz 2CPU (4cores on 1CPU, so all 8 cores) and on AthlonX2 BE-2400 2.2GHz (2cores on 1CPU, so all 2cores) respectively.
CrazyStone
Crazy Stone, running onIntel Xeon, 8 cores @ 2.83 GHz
FirstGoBot
running on a Pentium 4, 2.67Ghz (single processor)
GNU
GNU Go, running on one core of a dual core AMD Athlon 64 processor running at 2.2 GHz.
HBotSVN
AMD Athlon(tm) 64 X2 Dual Core Processor 5000+
HouseBot
AMD Athlon(tm) 64 X2 Dual Core Processor 5000+
LeelaBot
Leela, running on Intel Core 2 Quad Q6600 at 2.4Ghz
ManyFaces1
was running on one CPU of a 2.33 GHz core2 Duo
ManyFaces2
was running on one CPU of a 2.0 GHz Core2 Duo
MonteGNU
GNU Go with UCT enhancements, running on one core of a dual core AMD Athlon 64 processor running at 2.2 GHz.
Orego
running on a Mac using two 3 GHz Dual-Core Intel Xeons (hence 4 cores altogether).
SimpleBot
running on one cpu of a dual core AMD Athlon.
StoneCrazy
Crazy Stone, running onIntel Xeon, 8 cores @ 2.83 GHz
WeakBot50k
running on one cpu of a dual core AMD Athlon.
wei2go
Wei 2 Go, running on MacBook 2.0 GHz dual core Intel, 1 Gb DDR2 RAM