[Top][All Lists]

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

Re: [gnugo-devel] 261 games 3.1.31 vs 3.1.32

From: Gunnar Farneback
Subject: Re: [gnugo-devel] 261 games 3.1.31 vs 3.1.32
Date: Tue, 23 Apr 2002 17:43:53 +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)

Dan wrote:
> This game shows that GNU Go is not good at evaluating
> the weakness of groups. Of course, we know this from the
> NNGS games.

This is one of the areas where we have potential for a substantial
strength improvement. I have two reasonably simple ideas we could try
after 3.2.

1. One of the measures used when assessing group strength is the size
of the surrounding moyo. When evaluating the effectiveness of a
strategical attack or defense, it should be useful to take into
account how the size of the surrounding moyo changes with the move. As
it happens we already have this information available, so we just need
to find a good way to integrate it with the move valuation.

2. I believe that the owl code would benefit from a major redesign,
now that we have gained experience of this kind of reading, but that
is not quickly done. A simpler experiment would be to run the owl code
twice, where the first round has the restriction that only critical
moves and eye-making patterns may be used. Groups which can live under
this restriction would count as strong relative to ones which have to
run. The main work here would be to go through the patterns and decide
which ones have a definite eye-making purpose.

I have enough to do with other aspects of the engine, primarily
connections, so I'd be happy to leave this for other people. Possibly
Arend already is planning to do something similar to my first idea.


reply via email to

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