[gnugo-devel] SlugGo and twinbot (was Fwd: [computer-go] 12th KGS )

[David G Doshay]

I read your Senseis wiki page too, and see that we are doing similar  
things.
But because SlugGo runs on  cluster, our white and black evaluations are
done at the same time. Thus, we don't pay a time penalty for that.

It looks like you are using GNU Go specific evaluations for when you use
and when you ignore the opponent's choices. We use GNU Go more as a
black box generating ranked moves and then layer other evaluations on
top.

One thing I am seeing about what SlugGo does is that it is a little too
greedy about stealing the opponent's good move, so I need to adjust
some weighting factors. You seem more concerned about use for
defensive reasons, but in either case SlugGo does global look ahead
on a collective set of moves from both the white and black genmove(),
so we find out something about the consequences of those moves.
It is in doing these look aheads that SlugGo pays the time penalty; even
though they all happen in parallel, they are each serial in nature.

It seems likely that the ICOB paper will be online soon.

Cheers,
David



On 4, Mar 2006, at 2:57 AM, alain Baeckeroot wrote:

> I read tonight your description of SlugGo on senseis.net, and  
> discovered that
> my patch is based on the same go proverb : opponent good move is  
> (often) my
> good move. So i m very curious too see how SlugGo is, and hope to  
> learn a lot
> from it.