[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] owl tuning question
From: |
Evan Berggren Daniel |
Subject: |
Re: [gnugo-devel] owl tuning question |
Date: |
Sun, 8 Sep 2002 13:03:05 -0400 (EDT) |
On Sun, 8 Sep 2002, Arend Bayer wrote:
> On Sat, 7 Sep 2002, Evan Berggren Daniel wrote:
>
> > I was looking at a problem found by matcher_check (test 14 in
> > auto_nngs.tst), and I don't see why owl_attack and owl_does_attack fail to
> > find the attack on E10 of F11. dragon_status shows an illegal transition
> > later on when F11 is played by black.
>
> Probably that is a connection problem. I didn't check, but I think the
> following happens (I hope I get coordinates right, I am writing this
> from memory): GNU Go sees E10 is connected to G12. And it sees that G12
> is connected to G14. Hence E10-G12-G14 is considered as a single dragon.
> Which is obviously incorrect.
> Also, as it is, owl doesn't care whether a tail gets lost, as long as the
> dragon gets two eyes somewhere. So owl may just defend G12 and G14
> after black plays F11, and hence F11 is not an owl attack.
> After black plays F11, dragon amalgamation may be different.
> I am actually not sure whether there is a good way to make a test out of
> this. "same_dragon E10 G14" might make sense.
OK, I'll play with that a bit.
>
> > Also, is there a gtp command to give the reasons for a move? should there
> > be?
> No there is none.
> Hmm, I am not sure. The semantics of move reasons can change frequently,
> most of them are specific to GNU Go and they probably require
> human interpretation; and of course, they can be seen with -t.
I just like using gtp manually for looking at statuses when writing tests.
Thanks
Evan Daniel