[Top][All Lists]

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

Re: [gnugo-devel] owl problem 21

From: Daniel Bump
Subject: Re: [gnugo-devel] owl problem 21
Date: Fri, 12 Oct 2001 08:21:11 -0700

> Looks to me like D2 will also defend the B3 group.  It's less
> optimal, but does live.  Is the owl code expected to get the
> optimal move, or just any move?

If multiple defenses work, the regression should list
the acceptable ones, e.g. in owl.tst:

# incident 133
loadsgf games/incident121.sgf 122
22 owl_attack P12
#? [1 (PASS|S11|S14|N13)]
23 owl_defend P12
#? [0]

> >From a regression point of view, just any working move is okay. In
> practice it's a bit more complex. Some effort is made to find
> alternative owl attacks/defenses to choose from, but the chances that
> GNU Go plays the right move are definitely improved if the owl code
> comes up with a good move to begin with.

Right, the move found by the owl code has a good chance of
being the move that's actually made.

Another thing to be aware of, if the dragon is owl dead or owl
alive, the owl threat functions are run to determine if it can
be killed or made alive in two moves. The two moves that attack
are then cached in dragon.attack and dragon.second_attack.  I
think this potentially useful information that we do use now
and could make more use of. The engine would be faster if we
didn't do it (which we don't below level 8).


reply via email to

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