Re: computer-go: PN Search and transposition

["Antoine de Maricourt" <antoine.demaricourt@xxxxxxxxxx>]

> 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.

Yes, this is what I suspect as well, and I was starting to look towards this
direction.

> 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".

Yes. That's basically what he says. But I have no idea of what properties
this method has.

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

I wouldn't bet anything on this...

Antoine.