Re: [computer-go] An [open] question on game tree search theory

["Vlad Dumitrescu" <vlad_dumitrescu@xxxxxxxxxxx>]

Hi,

From: "Antti Huima" <antti.huima@xxxxxxxxxxxxxxxxx>
> I wish to share a problem I've been lately thinking of.

And it's a very interesting one too! :-)

> The question is: what structure the evaluation function E must have so
> that it is possible to argument that a min-maxified value of E upto a
> search depth is a better guide for game play than the flat evaluation of
> the successor positions by E itself? Why min-maxify E?

IMHO, a sufficient condition for this to work is that E(t) becomes a better
approximation for G(t) when t gets closer to the bottom of the tree. This also
means that at any leaf node, E(t) and G(t) should give exactly the same answer
to the question "who won?".

The more consistent E(t) and G(t) are, the more a minimax search will help.
This could also be used to test E(t): if it's affected a lot by a deepening
minimax search, then it isn't very good.

An interesting practical use would be to optimize the cost of computing E
against it's efficiency (the cost of searching deeper). Does anyone do this?

regards,
Vlad
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/