Re: computer-go: Evaluating positions

[Heikki Levanto <heikki@xxxxxx>]

On Mon, Jun 18, 2001 at 01:47:02PM +0200, Rafael Garcia Leiva wrote:
> Also I have a question. If the tree-branching-factor of a 9x9 go board
> is simmilar to tree-branching-factor of chess, why the go programs do
> not play better in a 9x9 board? Is it a problem of the evaluation
> function? Why?

Yes, the evaluation function is a big part of the reason. There is no easy
way to analyze if a group is alive or dead. Getting it wrong witll
completely spoil any other criteria you may use for evaluation. Sometimes
the differences are subtle indeed.

Another problem is that there can be quite long local sequences even in 9x9
go, for example a ladder across the board. Those need to be read out to the
end, there is no way to estimate the result halfway.

Finally, I think the branching factor for even 9x9 go is more than for chess
- there are 81 possible opening moves! And the game may well last longer...

Just my $0.02

-H


-- 
Heikki Levanto  LSD - Levanto Software Development   <heikki@xxxxxxxxxxxxxxxxx>