[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] attack_either testcase
From: |
Evan Berggren Daniel |
Subject: |
Re: [gnugo-devel] attack_either testcase |
Date: |
Sun, 24 Nov 2002 14:44:03 -0500 (EST) |
On Sun, 24 Nov 2002, Arend Bayer wrote:
>
> Evan wrote:
>
> > This is a new attack_either testcase. It is from the constraint on CB328
> > in nngs3:800, which should match but does not in current cvs.
>
> Evan, this looks like another important improvement due to your patch
> (I assume this passes with your patch?). However, in all cases I've
> seen, the PASSes were caused by the revisions in attack_either. Maybe
> you can try again whether we get a better speed/correctness trade-off
> if you leave out the changes to defend_both?
Yes, my patch fixes this.
I tried only applying the attack_either portion of the patch. Running
only against tactics1, it only gets the PASSes at tests 101 and 104, while
the full patch gets a PASS at 102 as well. I'll try running the full
suite with just the attack_either portion. Of course, part of the reason
that the defend_both patch helps is that a better defend_both means that
attack_either will do better, since they call each other recursively.
>
> [I've just merged this one and your endgame test cases.]
>
> Arend
>
> > +loadsgf games/nngs/Lazarus-gnugo-3.2-200205011927.sgf 86
> > +trymove black G4
> > +trymove white G3
> > +103 attack_either E5 G3
> > +#? [1]
> > +104 attack_either G3 E5
> > +#? [1]*
> > +popgo
> > +popgo
>
> And in case someone wonders, yes, with attack_either as in CVS, the
> order of E5 and G3 matters. (This should be solved by Evan's patch.)
Could this be a source of rotational dependence?
Thanks
Evan Daniel