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

["Mark Boon" <tesuji@xxxxxxxxx>]

For a while I thought this too, but actually it's flawed. This is because it
doesn't account for the loss of liberties caused by playing on the edge.
Therefore influence near the edge should be higher than what you get by
mirroring in the edge.

And then there's the matter of inefficiency of course of calculating
influence which is then discarded.

> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx
> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx]On Behalf Of Tom Cooper
> Sent: Monday, November 01, 2004 11:17
> To: computer-go
> Subject: Re: [computer-go] Designing faster, better influence functions
>
>
> I'm not sure how obvious the 'method of images' from physics would be to
> someone who
> hasn't come across it before.  It is probably worth considering, so I'll
> describe it
> in case it isn't obvious.
>
> The idea here would be to make a larger board, extending beyond
> the edge of
> the true
> board, and populate the extra regions with reflections, in
> mirrors situated
> at the edges
> of the true board, of stones on the true board.  Run the
> influence function
> for the larger
> board, but throw away the results that don't lie on the true board.
>
> At 18:11 31/10/2004, you wrote:
>
> >From: "David Fotland" <fotland@xxxxxxxxxxxxxxxxx>
> >
> > > Bouncing off the edge of the board is important, unless you
> compensate for
> > > the edge some other way.
> >
> >
> >OK, so I will use modified kernels near the edge.
> >
> >_______________________________________________
> >computer-go mailing list
> >computer-go@xxxxxxxxxxxxxxxxx
> >http://www.computer-go.org/mailman/listinfo/computer-go/
> >
> >
> >
> >---
> >Incoming mail is certified Virus Free.
> >Checked by AVG anti-virus system (http://www.grisoft.com).
> >Version: 6.0.778 / Virus Database: 525 - Release Date: 15/10/2004
>

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