computer-go: How to evaluate unbalanced alphabeta tree?

["Mousheng Xu" <moushengxu@xxxxxxxxxxx>]

I am wondering how to evaluate the minmax tree with alphabeta pruning when 
the subtrees are not of the equal size. For instance, when you have such a 
tree:

                    B(3,3)
                    /     \
                W(3,4)   W(4,3)
                /    \
             B(4,4) B(4,5)

Note the subtrees of B(3,3) are of different depth. This is what I call 
unbalanced.

It seems to me that in Go, it's reasonable to stop arbituarily in the tree 
sometimes. One example is when a large group of stones are captured, then 
you don't want to consider further. Another example is when you have run out 
of patterns at some branches. Thus the minmax tree for Go can be reasonably 
unbalanced.

The problem is that when the branches are unbalanced, you maynot always have 
a fair comparison of the estimated value. For the example given above, Black 
has one more play in the left branch than in the right, and this is 
advantageous to Black. High values at the left leaves does not mean W(3,4) 
is a worse choice than W(4,3).

What do you guys do with this problem?

Thanks.

-- Mousheng Xu


_________________________________________________________________
MSN Photos is the easiest way to share and print your photos: 
http://photos.msn.com/support/worldwide.aspx