gnugo-devel
[Top][All Lists]
Advanced

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

Re: [gnugo-devel] Gnu Go 3.2 possibly correctable flaw in play


From: SP LEE
Subject: Re: [gnugo-devel] Gnu Go 3.2 possibly correctable flaw in play
Date: Fri, 25 Jul 2003 13:27:57 -0700

>
> > On Fri, 25 Jul 2003 address@hidden wrote:
> >
> >> I noticed a possibly correctable flaw in Gnu Go 3.2 opening
strategy.
> >>  The
> >> SGF file below shows the problem with Gnu Go's opening.  I can use
> >> the same flaw
> >> to give Gnu Go 6 stones and still win, even though I'm not that
good.
> >>  This
> >> trick of sacraficing a couple of stones to waste a lot of the
> >> programs stones
> >> works against Swiss Explorer and Goliath also, but doesn't work
> >> against human
> >> players.  This looks to me like a straightforward flaw that might
not
> >> be too
> >> hard to fix.
> >
> > Thanks for pointing this out; it is indeed a problem.  The latest
> > development versions are less susceptible to this, but only
somewhat.
> > You
> > might try gnugo 3.3.23 for a stronger version; alternately, 3.4
should
> > be
> > out in a couple weeks.
> > I'm not sure how easy to fix this will be, but it definitely needs
> > attention; thank you for the report.
> > Evan Daniel
>
>
> Version 3.3.23 with the cosmic patch seems to solve the problem
> in that particular game, because it puts more emphasis on creating
> moyos in the beginning of the game. Look for instance at the moves
> it proposes between 6 and 18 :
>
> Move 2 (white): GNU Go plays D16 (75.00) - Game move D16 (75.00)
> Move 4 (white): GNU Go plays Q5 (25.08) - Game move Q16
> Move 6 (white): GNU Go plays R5 (44.98) - Game move D17 (23.21)
> Move 8 (white): GNU Go plays C5 (40.41) - Game move Q17 (23.89)
> Move 10 (white): GNU Go plays C5 (45.61) - Game move P18 (19.58)
> Move 12 (white): GNU Go plays E12 (33.23) - Game move P16 (31.86)
> Move 14 (white): GNU Go plays C12 (30.65) - Game move E18 (17.16)
> Move 16 (white): GNU Go plays R12 (31.85) - Game move E16 (22.71)
> Move 18 (white): GNU Go plays F12 (30.58) - Game move F17 (23.83)
> Move 20 (white): GNU Go plays M17 (36.25) - Game move G18 (23.33)
> Move 22 (white): GNU Go plays O12 (32.12) - Game move C12 (29.02)
> Move 24 (white): GNU Go plays D12 (45.60) - Game move D12 (45.60)
> Move 26 (white): GNU Go plays H11 (41.21) - Game move H17
> Move 28 (white): GNU Go plays F11 (38.01) - Game move F11 (38.01)
> Move 30 (white): GNU Go plays G10 (48.98) - Game move G10 (48.98)
> [..]
>
>
> Stephane.
>
>
>

I think it's certainly not easy to fix this problem, because there are
hardly clear cut moves which could be said as the only correct solution
at various stages of the game. One thing is sure, gnugo doesn't know
when and how to suppress a moyo. The cosmic patch seems to concentrate
on building moyo, but not suppressing a moyo. Actually at move 12, the
black moyo is already too huge and has to be suppressed at the position
like K3. The moves gnugo played may be less efficient, but not
worthless. Gnugo has to know that the moves like K3 is larger than
anything else.

SP Lee





reply via email to

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