Ninety-sixth KGS Computer Go Tournament

In the table above,
   0 is a loss
   1 is a win
   J is jigo
   left superscript is the player's colour
   right superscript is the round in which the game was played
   a subscript shows how the result was determined:
      R for resignation
      T for time
      F for forfeit
      a number for the points difference after counting.
All the 0s, 1s and Js are links to the game record.

Black won 63 games, White won 87 games, there were 8 jigoes, and pachi forfeited two games by not playing in them.

We welcomed a new player to this event, DolBaram. It had only started playing on KGS on the previous day. DolBaram is the work of Korean programmer Lim Jaebum, whose program Baduki competed in the FOST cup in 1998.

In all, seven players registered for the tournament. I therefore added GNU Go, running on a single processor of my own desktop PC, to make the numbers even.

Results

In round 1, pachi did not join its game with gnugo3pt8, and eventually lost by forfeit when its time ran out. This happened because its operator had forgotten to set "mode=tournament" in its initialisation file. So it would play games against human users of KGS, and when one of these games ended, it would join a tournament game if one had started, otherwise play against another human.

Pachi continued to play like this until round 22, when its operator realised what was happening and restarted it with a corrrect initlialisation file. It only lost two games by forfeit (both against gnugo3pt8), but it joined many others late, giving itself a disadvantage. For example, in round 6 it joined its game against fuego9 with only six seconds of main time left; it still managed to get a jigo.

Round 8 gave AyaMC its first result which was not a win. It got a jigo against fuego9.

Round 9 gave AyaMC its first loss, to DolBaram.


DolBaram vs gnugo3pt8
Moves 47 - 61

I know nothing about how DolBaram works, but I assume it uses Monte Carlo methods. However it does not try to win by as little as possible, like MoGo and some other Monte Carlo programs. The figure to the right shows the end of its round 16 game with gnugo3pt8. With move 47, it could have made it very clear that it owns two thirds of the board; instead it went on to kill all of its opponent's stones.


AyaMC vs gnugo3pt8
Move 45

In round 34, AyaMC played the move shown to the left, against gnugo3pt8. I assume it chose it because it is as good as any other move. But it happens also to be a move that a human might play in this position, meaning "I am so confident that I have won this game, that I can afford to make this obviously worthless move. Why don't you count, and then resign?"

Annual points

Players receive points for the 2013 Annual KGS Bot Championship as follows:

DolBaram	8
Aya	5
Coldmilk	3
fuego	2
oakfoam	1

Details of processor numbers, power, etc.

AyaMC: Aya, MC version, running on 2 machines: 980X 3.3GHz 6cores + W3680 3.3GHz 6cores
Coldmilk9: Coldmilk, running on a 16-core Xeon, 2.9GHz.
DolBaram: DolBaram, running on an i7-970 (4 cores).
fuego9: fuego, running on a 12 core Intel(R) Xeon(R) CPU X5670 @ 2.93GHz.
gnugo3pt8: GNU Go version 3.8, running at its "level 19", with Monte-Carlo enabled, on a single 3.3GHz Intel i5-2500 CPU.
NiceGo19N: oakfoam, running on a mini cluster: i7-2600K + i7-920
Orego12: Orego, running on five nodes of our custom Linux cluster Fido. The node has two AMD Six Core Opteron 2427 2.2 GHz processors (60 cores total), 8 GB RAM, Centos Linux.
pachi: pachi, running on 2x Opteron 6134 (15 threads) with 64GiB RAM.