RE: [computer-go] Designing faster, better influence functions

["David Fotland" <fotland@xxxxxxxxxxxxxxx>]

I can't find the detailed description.  The basic idea is:

Start at each stone and expand to liberties, liberties of liberties, etc.
Don't
expand through stones or liberties of stones.  The influence of each stone
falls
off with taxicab distance as K*((2/3)^distance), and is summed at each
point.  Use separate sums for
white and black.  The K value depends on the strength of the group (1.0 for
alive, -1.0
for dead, other values for intermediate strengths).  If expansion runs into
the edge of 
the board, bounce off it, and continue expanding back into the board.

After expansion, compare the values of white and black.  Clamp the peak
values for white and
black to a threshold (about 2) indicating one full point of territory.  If
white and black are both
above a small threshold (about 0.3), territory is zero.  Otherwise the
territory is simple function 
of the two values, similar to larger-2*smaller.

I have found that the distance function is a little to slow at large
distances, so it falls off faster when distance
is 6 or more.  The actual current values are:

static float rtdist[] = {  /* falloff in influence with distance */
   0.f, 1.f, .66666f, .44444f, .30f, .16f, .10f, 
   .0600f, .04000f, .020f, .01f, .005f, .0f, .0f, .0f 
};

David

> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx 
> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx] On Behalf Of 
> Myriam Abramson
> Sent: Saturday, October 30, 2004 2:15 PM
> To: computer-go
> Subject: Re: [computer-go] Designing faster, better influence 
> functions
> 
> 
> 
> > I've described my algorithm before so no details now.
> 
> I'm sorry I missed it. Would you mind repeating it? Thanks. 
> 
> 
> -- 
>                                    myriam 
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx 
> http://www.computer-go.org/mailman/listinfo/computer-go/
> 


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