Re: computer-go: Re: alternatives to Alpha-Beta Search

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

At 09:46 AM 11/18/99 +0000, P.J.Leonard wrote:
> Incidently my experience of watching my PN engine attempt to solve
>problems
>is that it tends to stay in a local situation for several moves (until
>it realizes
>that there is another area that is less effort to prove). It is my
>strong
>belief that the PN scheme is closer to the way us humans think than
>alpha-beta.
>
>  cheers Paul.

I agree that some kind of best first search like PN is better than
alpha-beta, at least for those go searches of variable depth where
there are some simple forced lines and other huge subtrees that
go nowhere.  For life and death or string tactics the goal
is a simple binary (live or die, capture or escape, etc).  I don't
use PN because its estimate of the work to do to prove is just based
on the number of moves generated.  In go, you can always generate
a few hundred moves if you like :)  I use a heuristic to estimate the
amount of work rather than the number of moves generated and it seems to
work pretty well.

David