Re: computer-go: PN Search and transposition

[Martin Mueller <mmueller@xxxxxxxxxxxxxx>]

Note that PSS is also an approximation algorithm. My paper shows some
contrived examples where it behaves really badly. I suspect that real proof
numbers for DAGs can be computed faster than with Schijf's method, using
some kind of fancy branch and bound scheme, but that it would still be too
expensive to use in most applications.
As a starter, you can look at the following decision problem: Is the proof
number of the root equal to one, or larger than one. Can you do better than
trying to set all the leaf nodes to 'solved' one by one and backing up?

I don't remember anything about Schijf's "practical" method, unless it just
means "propagate numbers from children to (all) parents using the same
backup rules as in trees". But it's been a while since I had time to look
into such things. Ah, the busy life of a professor :(

>http://hoggy.virtualave.net/y.bishop/page1.htm

Looks like a nice collection of papers there. I just hope they're
respecting all the copyrights...

	Martin