Re: [gnugo-devel] "local" Go games

[Eric]

--- Evan Daniel <address@hidden> wrote:

> On Sun, 12 Sep 2004 18:19:38 -0700 (PDT), Eric
> <address@hidden> wrote:
> > --- Evan Daniel <address@hidden> wrote:
> > 
> > > No, there's a difference.  If you cut down the
> board
> > > too much, you
> > > can't solve the local game.  You can, however,
> find
> > > moves that tend to
> > > be worth considering from a global perspective,
> or
> > > semi-local
> > > perspective (eg owl reading), even though you
> can't
> > > get definitive
> > > answers.  If, however, you cut down the game
> only
> > > somewhat, you can
> > > get correct answers (in some cases, anyway, like
> > > some endgame
> > > positions).
> > 
> > Yes, Evan, but you still have not yet defined "too
> > much" and "only somewhat" ;-)
> 
> Too much is when you start getting wrong answers,
> only somewhat is
> when you don't :)

Yes, this is what I said a while back - it is required
to implement and test in order to know the answer to
this.

> To be more helpful...  Owl patterns are in the "too
> much" category,
> and so larger-scale reading is needed.  For an
> example of "only
> somewhat" you might look at the active areas in the
> persistent caches
> (though frequently those are a bit on the aggressive
> side...  usually
> they work anyway though, but sometimes there are
> problems).

Yes, in the context of the current method of operation
of GNU Go.

> The real answer is that it is a non-trivial problem,
> and a good,
> consistent, correct solution would be a considerable
> contribution to
> computer go in general, not just GNU Go.

Okay.