[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] Designing faster, better influence functions
Frank wrote:
> A simple loop would do.
> Experimenting with values between Pyth. & Manh. and doing self-play.
> Perhaps someone using the GNU Go sources could test it.
GNU Go's influence function is highly tuned and enhanced by patterns.
Changing it to something else is likely to be very detrimental.
The shape of the influence function can be modified to some extent by
changing the value of DIAGONAL_DAMPING in engine/influence.h.
> But when GNU Go uses Bouzy's 5/21 it might not be so easy to modify.
GNU Go hasn't been using a Bouzy influence function in the move
generation for more than four years. (It has remained in one of the
scoring algorithms, however.)
/Gunnar
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/