|
Formal division: 13x13 board Chinese rules, komi 7½ 43 minutes each absolute time |
Open division: 13x13 board Chinese rules, komi 7½ 28 minutes each absolute time |
Formal, four-round Swiss; Open, six-round Swiss.
The first round started at 08:00 UTC for the Formal and 08:10 for the Open division, subsequent rounds started at ninety-minute and at one-hour intervals for the two divisions.
As usual, the tournament was held in two divisions, Formal and Open, with more restrictive entry conditions for the Formal division.
Formal Division 13x13
|
Open Division 9x9
|
The "real" names of the bots listed above, and of their programmers, are listed here: programs which have registered for KGS Computer Go Tournaments.
The small boards, and slow time limits, of this event made it suitable for programs using Monte Carlo methods. These I know include Crazy Stone (playing as CrazyStone and as StoneCrazy) and MoGo (playing as MoGoBot13 and as MoGoBot). MoGo uses a "bandit" Monte Carlo method, as described at http://zaphod.aml.sztaki.hu/papers/ecml06.pdf.
My impression of MC bots has been that they play games which are rather dull to watch. Some programs, HandTalk in particular comes to mind, give the impression that they "know what they are trying to do", even if their execution of it is defective. This tends to produce games that are entertaining to watch. However a MC bot is not trying to accomplish anything, it is just trying to make good moves, and I have felt the result, though effective, was dull for the human observer (but then, I am the kind of observer who enjoys watching TheCaptain's game on KGS).
However, I have been forced to change this opinion, after watching games like the one in the diagram below.
Viking5 was beaten only by CrazyStone, CrazyStone was beaten only by MoGoBot13, GNU was beaten only by viking5. Viking5 was ahead on SOS (despite getting a bye in the final round), and so was declared the winner. Unlike the MC-based MoGo and Crazy Stone, viking5 is based on pattern-matching.
In round 3, MoGoBot13 showed its strength, and weakness, in its
game with viking5. In the position shown,
its moves from 1 to 9 appeared to be killing the black group competently (in fact
it started its attack several moves earlier). However, with move 11, it abandoned
the killing and let the group live, losing the game as a result. Its author Yizao
Wang attributed this to a pattern bug in MoGo.
In round 1, HouseBot beat WeakBot50k, the whole game taking 24 seconds.
In round 6, when the IdiotBot/MoGoBot game started, MoGoBot did not join
it. After a while IdiotBot left. Then MoGoBot joined it, made a move, and eventually
won on time.
Incidents like this give me, as director, a slight ethical problem: how far should I
go in reminding the bot's owner that it needs attention? The policy I follow is to
take "reasonable steps" to inform the bot's owner if it fails to join a game. If the
owner is on KGS, I shall try to mention it to them. If they are not, I shall probably
do nothing.
The following "forfeited game" record is available by clicking on it:
MoGoBot-WeakBot50k.sgf
Other game records are available from the KGS pages: Formal Division. Open Division.