[Top][All Lists]

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

Re: [gnugo-devel] trevor_1_15.2

From: Gunnar Farneback
Subject: Re: [gnugo-devel] trevor_1_15.2
Date: Tue, 20 Nov 2001 22:34:09 +0100
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)

Trevor wrote:
>  - removed  "strategic effect".
> The strategic effect seems to cause almost as many problems as it solves.  
> Here's a list of FAILURES (35) / PASSES (26) with removing the effect.
> There are a lot of fixmes in this section of code.  Is this effect 
> really necessary?  If so, it probably needs an overhaul.

It certainly has weaknesses, but yes it's definitely needed. Without
it you'll for example get no value (except an occasional point of
territory) for cutting or connecting groups unless the cut or connect
does kill or save one of the groups.

It may be worth reminding that the regressions almost only cover
positions where GNU Go has made mistakes in the past. Thus there are
many examples where the weaknesses in the strategical evaluation cause
bad moves, but very few examples where the strengths of the
strategical valuation cause good moves.

It's not really statistically significant, but it may be an indication
that global.tst has 5 failures against one pass. Those positions are
not from GNU Go games but from Golois games and thus less biased in
this respect.

> I'm currently updating the regression view every couple of days
> with the latest CVS.  If anyone shouts, I'll add the timestamp
> to the html display, so you can tell when it was built.

A timestamp (ideally for the "cvs update", but the build time should
be close enough) may turn out useful sooner or later.


reply via email to

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