Re: [gnugo-devel] Anomalous Game

[Paul Pogonyshev]

Gunnar wrote:
> Paul wrote:
> > >  Pattern 448
> > >
> > > - !
> > > -!*x
> > > + )
> > > +)*x
> > >
> > >  :0011
> >
> > This patch has a bad regression delta:
> >
> > [...]
> >
> > I still think this patch is correct.  I think we should accept it
> > and improve vital point selection separately.
>
> Yes, it looks perfectly okay. It may still be worth taking a closer
> look at one of the failures though.

nicklas1:1213   Accidential fail.  GNU Go now sees that the eye at K8 is
                critical with a vital point at L9 (improvement!).
                However, this causes a semeai node overflow in semeai of
                F15 against H12.

owl1:293        This one is not so clear, but still seems an improvement
                to me.  Consider variation after W S2, B S6, W S4, B S1,
                W S3.  Patched GNU Go thinks that Q5 is vital for the
                eye around R5.  If you don't take into account all the
                possible ataris around this eye and the fact that white
                P5 is probably sente (GNU Go can't read such things at
                present) then yes, the only attacks against the eye are
                Q5 and S5 (the latter works for a "wrong" reason, though,
                but that's not something optics code can understand) --
                exactly as read_eye() tells.

manyfaces1:30   No idea.  But it's a pass :)

Paul