Re: [gnugo-devel] News from the olympiad

[jan . villemson]

On Sun, 4 Jul 2004, Gunnar Farnebäck wrote:

> I notice that the komi is wrong in the game records saved by GNU Go,
> so you probably are missing to give the "--komi 6.5" command line
> option when starting GNU Go. I might add that setting the komi to 6.5
> in Cgoban only affects the scoring done by Cgoban but is not
> transfered to GNU Go. In fact it is not even possible to send komi
> over GMP; the protocol has no support for it.

OK, I was not aware of that. Still, I think this hasn't affected any of
the outcomes.

> I also see that timing information only is available in
> mag-gnu-9-c.sgf. Presumably this is caused by the advice to set the
> timing system in Cgoban to none (we don't want to risk Cgoban killing
> the game prematurely). Maybe it is better to use absolute time but
> with double the time or so, for margin.

Anyway, the timing information for GNUGo is not very interesting, it
usually moves very fast, spending no more than a minute or two on whole
9x9 game. The exception is a test game we played with Many Faces of Go,
the latter running the strongest available version and GNUGo using level
15. I restructured the webpage a little and you will find this game under
today's date.

So does this mean there is not much sense in saving two copies of each
game?

> Btw, how are the time limits
> enforced in practice in the Olympiad?

They are not really. This far we have had no problems with programs
exceeding the time limit, and even if someone did, the rules say that it
is enough to play 30/60 moves in a 9x9/19x19 game. But GNUGo is very-very
fast.

Right now the result in 9x9 contest is that GNUGo has lost 2 games to
GoIntellect and 1 game to Magog, Many Faces and DumbGo. There is one more
game to play against DumbGo and then we will know its final score.

Greetings from Israel,
Jan