[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [computer-go] Designing faster, better influence functions
I think we agree that to determine territory, dead groups need to be removed
before a final influence pass. MFGO also does a preliminary influence pass
to
find completely surrounded area for big eye space. But that has nothing to
do with
territory and uses a different algorithm. Small eyes are recognized
directly without
any influence pass.
As I said, his influence is fine depending on what it is used for. I took
his question to
be whether or not his gif file gives a reasonable territory estimate, and I
still think it
doesn't.
The lower right group is dead because it can't make two eyes. If it had two
more liberties it would
still be dead.
The left group has lots of potential to make eyes or run away if it moves
first, so I think it would
be a mistake for the static evaluation to consider it dead. I remove dead
groups
based on the static group evaluation, and tactical stability is only a small
part of
that evaluation.
I do a little more than Goliath does, since the amount of influence radiated
from a stone depends
on the strength of that stone. I don't just remove dead stones.
In any case, details of the influence function are not very relevant to
playing strength which depends
much more on the tactician and life/death/strength evaluation.
David
> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx
> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx] On Behalf Of Mark Boon
> Sent: Saturday, October 30, 2004 12:48 PM
> To: computer-go
> Subject: RE: [computer-go] Designing faster, better influence
> functions
>
>
> I also don't think this is a fair critique, calculating
> influence is something else completely from determining what
> is dead or alive. Why does ManyFaces consider the lower-right
> dead? Probably your 4-liberty tactician found out. What if it
> had one or two liberties more? Why is the white group on the
> left not considered dead?
>
> Goliath also 'removes' dead stones if the tactician finds out
> they can't escape, but it's limited to 2-liberty groups.
>
> It's a bit of a chicken-and-egg problem. Before you have a
> good influence you first need to know what's dead or alive.
> How to know what's dead? You need to look at eye-space (among
> other things), which is derived from influence...
>
> In Goliath I solved this by first computing the influence.
> From that I derive groups, territory, eye-space and other
> stuff that basically determines the strength of groups. Then
> I try to determine which groups are dead by looking at their
> (relative to neighbours) strength and eye-space. When
> considered dead, a group is 'removed' and the influence is
> calculated again for the score.
>
> > -----Original Message-----
> > From: computer-go-bounces@xxxxxxxxxxxxxxxxx
> > [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx]On Behalf Of Frank de
> > Groot
> > Sent: Saturday, October 30, 2004 20:50
> > To: computer-go
> > Subject: Re: [computer-go] Designing faster, better
> influence functions
> >
> >
> > From: "David Fotland" <fotland@xxxxxxxxxxxxxxxxx>
> >
> > >It doesn't look reasonable to me, since it gives territory to
> > dead groups.
> >
> > Bruno Bouzy's 5/21 function also does.
> > These influence functions are based on the fact that dead
> stones are
> > removed first.
> >
> > Because Bouzy in his thesis did not remove dead stones and
> I wanted to
> > use the same position for comparison, I also did not remove dead
> > stones.
> >
> >
> >
> > _______________________________________________
> > computer-go mailing list
> > computer-go@xxxxxxxxxxxxxxxxx
> > http://www.computer-go.org/mailman/listinfo/computer-go/
> >
>
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
>
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/