gnugo-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] inconsistnies in owl_attack and owl_does_attack


From: Gunnar Farneback
Subject: Re: [gnugo-devel] inconsistnies in owl_attack and owl_does_attack
Date: Sat, 17 Aug 2002 22:59:23 +0200
User-agent: EMH/1.14.1 SEMI/1.14.3 (Ushinoya) FLIM/1.14.2 (Yagi-Nishiguchi) APEL/10.3 Emacs/20.7 (sparc-sun-solaris2.7) (with unibyte mode)

Evan wrote:
> Really?  what we're seeing in the matcher_check results is that there are
> moves that are attacks that owl_attack fails to find, that are evaluated
> highly enough to be played even though they are not noticed as
> owl_attacks.  This is where the inconsistencies come from.  That suggests
> there might be some for the ones valued second or third that should be
> valued highly.  Maybe I'm missing something, but that's what I think the
> results mean.  Doesn't mean it's the right way to go, but we should either
> look closer or just try it and see.

Right, I hadn't thought through all consequences of all examples being
from self-play. That's indeed the only way to get inconsistencies in
that scenario. I'm fairly certain that the situation is quite
different in games against other players though, especially stronger
ones. Losses in handicap games (with GNU Go as black) is likely to be
a good class of game records to investigate.

/Gunnar




reply via email to

[Prev in Thread] Current Thread [Next in Thread]