Hundred and twenty-fourth KGS Computer Go Tournament

Sunday July 10th, 2016

These results also appear on an official KGS page.

Rules

formatsix-round Swiss
board size19×19
rulesChinese
komi
time29 minutes plus 10/30s

Times

The first round started at 16:00 UTC.

Result table

PlaceNamecross-tableWinsSOSSoDOSNotes
Zen19 AyaMC Leela ManyF
1Zen19X
X
W12R B15R B11R W16R B1321½ W14R 61212Winner
2AyaMC B02R W05R
X
W03R B14R B11R W1610½ 3184
3LeelaBot W01R B06R B13R W04R
X
W02R B15R 2204
4ManyFaces1 W0321½ B04R W01R B0610½ B12R W05R
X
1222

Black won 8 games and White won 4.

Players

Six players registered, four for the Formal division and two for the less demanding Open division.

Results

Play in the Formal division proceeded normally.

However, we were all surprised to see that the Open division had been set up to use 9×9 boards, but with the correct 29 minutes each. This was my fault. I apologise, particularly to Katsuki Ohto, the creator of JulieBot, and to Gonçalo Mendes Ferreira, the creator of matilda.

JulieBot was unaware that the boardsize was different from what it expected, and tried to play at q16, which was rejected by the server. So play was impossible.

I cannot alter the setting of a KGS tournament once it has started. I could have created a new tournament starting a few minutes later with the correct settings; but with the 9×9 tournament still running, the bots would not have known which tournament to join. I cannot close the 9×9 tournament while it is still running. I could have removed both bots from the 9×9 tournament, but this would have left a tournament with fewer than two players, and I knew from previous experience that when the KGS tournament software needs to choose the pairings in a tournament with fewer than two players, it hangs, bringing all ongoing tournaments to a stop.

Annual points

Players receive points for the 2016 Annual KGS Bot Championship as follows:

Zen8
Aya5
Leela3
Many Faces of Go2

Details of processor numbers, power, etc.

AyaMC
Aya, running on a Xeon W3680 3.3GHz 6 cores.
LeelaBot
Leela, running on an Intel Core i5-6600 + AMD Radeon R9 390.
ManyFaces1
Many Faces of Go, running on a 4-core i7-4790 3.6 GHz, no GPU, using a deep neural net trained on KGS games.
Zen19X
Zen, running on a dual 10-core Xeon E5-2687W v3@3.1 GHz/32 GB RAM