Re: computer-go: Proof number search

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

It's an interesting algorithm, but not for Many Faces.  PN search works 
best when the
tree is irregular, perhaps with long forcing sequences, since it focuses 
the search
where there is less work to be done.  It assumes no knowledge about move 
ordering.

In go tactics, it is pretty easy to accurately order the moves, but its 
hard to absolutley
eliminate all but a few moves unless there is a ladder.  In a ladder it is 
trivial to find
the best move quickly, so there is no need for PN search.

If you don't want to spend the time to count potential new liberties or 
potential new eyes
to do move ordering, then PN search might help.  BUt if you are doing life 
and death without
recognizing eye shapes you have to read so much deeper you had better have 
a really fast
computer :)

Regards,

David

At 05:37 PM 5/14/2003 -0700, you wrote:
> I'm reading Dennis Breuker's thesis (thanks for the link, Erik), and it's 
> fascinating.
>
> This "proof number search" algorithm sounds like it would be fabulous for 
> ladders and life-and-death.  Is anybody using this for Go, or was I just 
> handed a piece of low-hanging fruit?  :-)
>
>
> Peter Drake
> Assistant Professor of Computer Science
> Lewis & Clark College
> http://www.lclark.edu/~drake/