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

["Jim O'Flaherty, Jr." <jim_oflaherty_jr@xxxxxxxxx>]

Frank,

Excellent question, I think.

I think it might make an interesting parameter to a genetic algorithm search.  I bet the value
would fall between these two calculations, Pythagorean and Manhattan.  And I bet the accuracy of
this value becomes more and more important as "influence" becomes more and more important in the
evaluation function(s) (as opposed to pattern matching and tactics).


Jim


--- Frank de Groot <frank@xxxxxxxxxxxxxxxxx> wrote:

> To: "'computer-go'" <computer-go@xxxxxxxxxxxxxxxxx>
> Sent: Sunday, October 31, 2004 7:04 PM
> Subject: RE: [computer-go] Designing faster, better influence functions
> 
> I was wondering, why do most influence functions (like Fotland's) use the
> Manhattan distance for the strength of the influence and not the Pythagorean
> distance?
> 
> I can see that the argument is that the Manhattan distance is the number of
> stones needed to connect, but in Go, diagonally connected stones are pretty
> strong, so one would expect that the Manhattan distance is mainly used
> because it speeds up the calculation perhaps?
> 
> _______________________________________________
> 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/