Re: computer-go: Learning from existing games

[Heikki Levanto <heikki@xxxxxx>]

On Tue, Jan 14, 2003 at 08:11:52PM +0100, schraudo@xxxxxxxxxxxxxxxxx wrote:
> I suspect that throwing away all spatial information will cost you, at
> least in that the program won't learn to play good shape.  Perhaps combine
> it with a local shape-based candidate move generator (which could also be
> a neural network)?

Well, if you want to add good shape, it is easy to add counters on how many
good-shape patterns match, and how many bad-shape patterns match, and let
the program learn if they have any value...

Lots of this will happen behind the scenes in the pre-analysis phase. For
example, the program may learn to avoid empty triangles - not because they
are ugly, but because they produce less liberties than other shapes.

But as I said, this all is empty speculation from my part - I haven't tried
any of it out. Maybe I'll have more spare time in the next decade...

-H


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