Re: [gnugo-devel] nando_3_16.5.tar.gz

[Gunnar Farneback]

Arend wrote:
> I've added this to CVS; I added a couple of alternative moves for some
> of the owl tests.

What about PASS as defense in nando:12? To me the dragon looks alive
as it stands. It can either make two eyes at the top or connect out
with K12.

> > +# Either the reverse followup or the constraint of EE106 is
> > +# wrong IMHO
> > +#CATEGORY=ENDGAME
> > +loadsgf games/nando/auto012.sgf 112
> > +16 gg_genmove white
> > +#? [!D1]
> 
> I agree that there are bigger moves on the board (e.g. M18), but the
> reverse followup seems correct here. Black C1 would be sente, and D1 is
> worth 6 pts.

Agreed, the valuation of D1 looks good. To me it seems strange to have
a test case which only requires a reasonably valued move not to be
played. Either remove the test case or list a couple of bigger moves
as correct answers.

Nando wrote:
> My point about the constraint of EE106 is that it only checks that an X
> move would be sente, but it doesn't check whether O can block on the first
> line or not. In situations like those described by the EB716 pattern for
> instance, I would definitely agree with that reverse followup value, but
> unless I missed something, the constraint of EE106 is different.

Huh, what does it matter whether O can block on the first line? The
constraint expects O to do that and then checks whether the cutting
point must be connected after X connects back. If O has to fall back
on the second line to begin with it's just a bonus.

(The exact size of the reverse followup value may be somewhat off
occasionally, but it shouldn't be all that critical.)

/Gunnar