gnugo-devel
[Top][All Lists]
Advanced

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

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


From: Eric
Subject: Re: [gnugo-devel] "local" Go games
Date: Sun, 12 Sep 2004 20:05:28 -0700 (PDT)

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

> On Sun, 12 Sep 2004 19:17:09 -0700 (PDT), Eric
> <address@hidden> wrote:
> > > Yes, Evan, but you still have not yet defined
> "too
> > > much" and "only somewhat" ;-)
> > 
> > And even if you could, your definition would
> depend
> > upon an algorithm - you know, how your program
> rates
> > all of the local board "candidate" moves; and
> whether
> > it just *picks* a global move from the candidates,
> or
> > *synthesizes* a global move from the candidates.
> There
> > are a lot of unknowns here.
> 
> My guess is that it would be best to say it does
> something in between
> -- there are a bunch of move generators, and it
> picks from among those
> moves.  Some of the generators are very localized,
> and others are less
> so.
> 
> Also, frequently it would be incorrect to attempt to
> "synthesize" a
> move -- it is rare that the globally correct move is
> not one of the
> locally correct moves.  Unless you mean something
> like evaluating the
> different moves suggested with the other evaluators
> -- that is done.

Here I was referring to an approach in which a formal
system, e.g. a planner, were to be used to generate
all of the local moves. In this case, it is possible
to (logically) infer global facts (moves) based upon
an (logical) analysis of the set of local facts
(moves).

Then, the result of all of this would be a logic-based
move, either picked or synthesized, that could be
subsequently evaluated against all of the
pattern-based moves that the classic GNU Go move
generators suggest.

So you see, pattern-based and logic-based approaches
can be integrated.





reply via email to

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