On 4/5/07, Don Dailey <[EMAIL PROTECTED]> wrote:
On Wed, 2007-04-04 at 23:35 +0200, Erik van der Werf wrote:
> Sylvain, could you run the same test on 7x7 to verify that there the
> 'correct' komi would be 9 (try 8.5 vs 9.5)?
I can already tell you what will happen, I have done similar tests.
At this point, the good UCT programs are so strong that if you
use 8.5 black wins a pretty high percentage and if you use 9.5
white wins a fairly high percentage. I don't remember my exact
results on this. I would still like to see Sylvain test this.
You seem to be more certain on this than Sylvain :-)
Let's see what he gets out with 'only' the 3k playouts...
On 5x5 it's just not testable, the UCT programs will win all games
as black. I think Lazarus playing really fast only lost 1
game as black using 24.5 komi out of hundreds.
Right, on 5x5 the opening move in the center is too powerful. One way
to make 5x5 more interesting for these kinds of experiments is to
forbid the first move in the center, but even then the possible
variations are still rather limited (but at least significantly deeper
and both sides can live).
Erik
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
