|
From: | David G Doshay |
Subject: | Re: [gnugo-devel] Fwd: [computer-go] April KGS on-line computer Go tournament |
Date: | Sat, 6 May 2006 13:22:05 -0700 |
On 3, May 2006, at 1:26 PM, David G Doshay wrote:
On 3, May 2006, at 12:50 PM, Evan Daniel wrote:I have observed this same effect in metamachine games with GoFigure --even numbers of branched plies do worse than odd numbers. I would expect something like 8-4-2 to play better than 8-8. I think the problem basically arises from which side you "force" to make mistakes that GNU Go thinks are good moves first. With 2 branching plies, SlugGo is the player making mistakes first.We did try 8-3-2, which played worse than l3 or 88, although it may not have converged yet. It played as slugGo_B32.
We ran a short experiment, so the results can only indicate, but we are not
absolutely sure, nor do we have any idea what the sigma might be. We changed the branching from the 8-8 that SlugGo_88 was using to 8-6-4,and used the old name so that we started from the old rating. Over the next few days we dropped about 120 points. This is close to my earlier observation that 100 point swings are not uncommon, but is sure does not look like it was
any help in getting stronger. Cheers, David
[Prev in Thread] | Current Thread | [Next in Thread] |