Re: [computer-go] Bouzy/Zobrist algorithms

[Erik van der Werf <E.vanderWerf@xxxxxxxxxxxxx>]

Peter Drake wrote:

> I'm trying to implement Bruno Bouzy's "mathematical morphology" 
> variation on Zobrist's influence function.  Some questions:
>
> 1) Are the values capped at 64?  It doesn't say so anywhere, but I 
> can't figure out why the stones would still be at 64 after 3 dilations 
> in Figure X2.  (The paper in question is "Mathematical morphology 
> applied to computer go," which is available from the Online Computer 
> Go Bibliography.)
AFAIK there is no specific reason, but yes he did cap the values. 
According to Bouzy there is no specific reason for the value 64 other 
than it being sufficiently large. (Zobrist btw used 50 and his dilation 
operator is slightly different from the one used by Bouzy. IIRC the 
difference, which occurs when both signs are adjacent and one has a 
majority, did not significantly change the performance.)

> 2) I don't understand why e = d(d-1).
Careful, it should be: e = d(d-1)+1

This becomes the unity operator for a single stone on a sufficiently 
large board. IOW a single stone in the center of a sufficiently large 
board doesn't enclose any territory. In practice it seems that you can 
use slightly larger or smaller values for e without significantly 
changing the performance. However, my experiments confirmed that d=5, 
e=21 is a good setting (slightly better then d=4,e=13).
BTW once you start to use knowledge of life and death the value for e 
tends to become mostly irrelevant...

Erik

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