Forty-second KGS Computer Go Tournament

Sunday September 14th 2008

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

Rules

 Formal divisionOpen division
format8-round Swiss8-round Swiss
board size19×1919×19
rulesChineseChinese
komi
time18 minutes absolute18 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
1stLeelaBot73128
2ndManyFaces163319
3rdCzechBot53114
4thAyaMC53012
5thFuego43310
6thvalkyria193349
7thGNU2300
8thFudoBot0340

Open Division   19×19

placenamewinsSOSSoDOS
1stManyFaces283434
2ndLeelaBot273022
3rdMonteGNU53714
4thAyaMC253313
5thProject4319
6thHbotSVN3325
7thWeakBot50k3315
8thbreak192335
9thIdiotBot2285
10thSimpleBot1312

We welcomed a new program 'Fudo', playing as FudoBot, written by Hideki Kato. He is already known to competitors as the author of 'GGMC Go', which has played in these events as ggmcbot. Fudo is based on the technical report on MoGo.

We also welcomed 'Project', a program by John Davies. Project is derived from his 'DumbBot', which played in the first few KGS bot tournaments, in 2005.

No entries were received from MoGo or from CrazyStone. I accepted an entry from CzechBot, a build of MoGo by Petr Baudiš, on the understanding that I would remove it if MoGo's authors entered their own version. I favour having strong players such as MoGo compete in these events, and I thank Petr Baudiš for making this possible.

I was again impressed by the high level of play, despite the fast time limits for 19×19 boards; and of dispute resolution. Unlike the early days of these tournaments, I did not need to intervene at all. Some games went into the dispute resolution phase at the end, but the disputes were all resolved without my involvement.

I am considering changing the timing system used for these events, so that instead of say 18 minutes absolute time, they will use 18 minutes plus 20 stones in 20 seconds Canadian overtime. I want to use something with a fairly sharp cutoff, so that the schedule will not be disrupted by long games. However I want to avoid the situation where a program claims the status of the groups correctly but then loses on time in the clean-up phase. Of course this change will only be useful to programs which can do the meaningless moves of the clean-up reasonably fast.
       Please let me know what you think about this proposed change. I prefer emails to the computer go mailing list, but if you don't want your view made public, to me at nick@maproom.co.uk.

You may have noticed that http://www.weddslist.com/kgs/index.html now hosts an advertisement. I apologise for this. However the payment I recieved seemed significant, and is some compensation for the time I spend on these events.

Formal division

AyaMC vs FudoBot
Position after move 134

In round 1, AyaMC and FudoBot achieved a difficult semeai in the top left, as shown to the left SGF. I say nothing about the correct status of this semeai, because, like the kibitzers, I do not know. Eventually Black made two eyes in the top left corner, leaving the white group dead; but White later killed all the black stones in the lower right, and won the game.

Fuego vs GNU
Move 22-27.

Also in round 1, Fuego took advantage of GNU's worst weakness SGF. GNU has a way of abandoning weak groups when they become loosely surrounded. This is seen in the diagram to the right. After move 25 GNU made no attempt to save the stones in the top left or to use their aji in any way, so White ended with a huge territory there. Admittedly, in the sequence shown, Black gets two good free moves on the outside; but these are not sufficient compensation for the lost group, and White's 22 and 24 also have their value on the outside. White eventually won this game.

In round 5, LeelaBot had its only loss, to valkyria SGF. This was an exciting game, which I enjoyed watching.

Fuego vs AyaMC
Move 225, marked.

Also in round five, the game between Fuego and AyaMC ended with a sudden resignation, in the position shown to the left SGF. This seemed like a human reaction: "those four black stones were dead, and now he's lived them in a seki, I never saw that coming". I would be interested to know whether the seki was in fact recognised in the players' readouts. I suppose what makes this resignation seem human is that recognising a position like this as seki only after it has happened is characteristic of strong kyu players like myself.

In round 6, Fudo ran in a ladder against LeelaBot, not seeing that it was going to be captured SGF.

FudoBot vs ManyFaces1
Moves 118-120, marked.

In round 7, ManyFaces1 had a very difficult ladder to read in its game against LeelaBot SGF. After move 127, the last move shown in the diagram to the right, White cannot save both groups. But I cannot blame ManyFaces for failing to foresee this ten moves earlier. If Leela foresaw the conlusion of this ladder whan it played 117 I am impressed (maybe I shouldn't be – perhaps, for an MC program, reading a messy ladder like this one is no harder than reading a simple ladder).

Fuego vs CzechBot (MoGo)
Moves 111-150, marked.

Another exciting game in round 7 was between Fuego and Czechbot SGF. After Black's move 111, shown to the left, Black's group at the bottom has three liberties, while White's group above it has two. However White narrowly managed to keep his group alive and win this long fight with the sequence to move 150, after which Black resigned. Throughout this fight, both players seemed to know what they were doing, and there were no notably stupid moves.

FudoBot vs ManyFaces1
Moves 118-120, marked.

Also in round 7, ManyFaces1 misread another, simple, ladder in its game against Fudobot, as shown to the right SGF. This seems unlike the "classical" version of Many Faces of Go. It is also unlike good UCT programs, which can reliably read ladders as short as this.

Open division

SimpleBot vs WeakBot50k
Moves 130 & 131, marked.

The round round 1 game between SimpleBot and WeakBot50k ended when SimpleBot wanted to make a move forbidden by the superko rule SGF. SimpleBot was not allowed to make its move, and timed out. The position is shown to the right: White wanted to recapture one stone.

Also in round 1, the game between MonteGNU and break19 stopped with disagreement about the status of break19's groups SGF. They successfully switched to cleanup mode, esablished the status of the groups, and ended the game successfully.

In round 4, LeelaBot2 lost its only game, to ManyFaces2 SGF.

Also in round 4, break19 achieved a won position against IdiotBot, but lost on time SGF.

In round 5, HBotSVN lost on time, in a won position, to IdiotBot SGF. HBotSVN does its own clean-up before passing, and it had almost finished this when it suffered a brief disconnection.

In round 7, break19 and Project went into clean-up mode on move 402 SGF. Break19 was ahead on the board, but this was irrelevant: it had only one second left, while Project had two seconds. Break19 won on time.

In round 8, break19 lost on time during the clean-up of its game with LeelaBot2 SGF.

 

Processor numbers, power, etc.

AyaMC
Aya, running on Xeon X5355 2.66GHz 2CPU (4cores on 1CPU, so all 8 cores)
AyaMC2
Aya, running on Opteron 252 2.6GHz 4CPU(1core on 1CPU, so all 4 cores)
break19
running on a single processor Intel(R) Celeron(R), 1.7Ghz
CzechBot
MoGo, running on 2 cores of a four-way Intel(R) Xeon(R) CPU 5130 @ 2GHz
FudoBot
running on four PCs connected through a normal Gigabit Ethernet LAN. Each PC has one Intel Core2Quad processor running at 3GHz. So, 16 cores in total.
Fuego
was running on an 8 core machine with Intel Xeon E5420 at 2.5 GHz.
GNU
GNU Go, running on one core of a dual core AMD Athlon 64 processor running at 2.2 GHz.
HBotSVN
One core of an Intel(R) Core(TM)2 Duo CPU T5450 @ 1.66GHz
IdiotBot
running on Linux, 4GB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+
LeelaBot
Leela, running on Intel Core 2 Q6600 2.4Ghz
LeelaBot2
Leela, running on Intel Mobile Core 2 P9500 2.53Ghz
ManyFaces1
running on 64 CPUs of a cluster of 3.17 GHz Xeon, 8 CPU cores to a node.
ManyFaces2
running on 32 CPUs of a cluster of 3.17 GHz Xeon, 8 CPU cores to a node.
MonteGNU
GNU Go with UCT enhancements, running on one core of a dual core AMD Athlon 64 processor running at 2.2 GHz.
Project
running on Intel(R) Pentium(R) 4 CPU 1.6GHz
SimpleBot
Linux, 4GB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+
valkyria
running on a single Pentium-M 1.4GHz processor
WeakBot50k
running on Linux, 4GB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+