[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] arend_1_14.6a
From: |
Daniel Bump |
Subject: |
Re: [gnugo-devel] arend_1_14.6a |
Date: |
Fri, 16 Nov 2001 16:32:26 -0800 |
> GnuGo chooses randomly between different moves of class jF (resp. tF).
> Moves of class j/t without class F (which frequently appears in patterns.db)
> no longer have a maximum value. Choice between different such moves is
> according to GnuGo's valuation as a secondary reason (which is valued
> sufficiently low so that it does not overrule shape tuning).
> A move of class jF is preferred to class j unless GnuGo values the latter
> to be worth more than 20 pts; similarly with t patterns.
This seems at first to contradict the comment:
######################################################################
#
# If the maximum values for a j or t patterns should NOT be enforced,
# add the F classification.
#
######################################################################
which would imply that patterns WITH the class F no longer have
an enforced maximum, the opposite convention as in your patch comment.
I guess the key is this comment:
/* Patterns in fuseki.db are assumed to be a fuseki pattern.
* The letter F toggles this.
*/
if (fusekidb) pattern[patno].class ^= CLASS_F;
This seems potentially confusing. Maybe better just manually add the
F class to all patterns where it is applicable. This shouldn't be
hard to do in fuseki.db using an emacs macro.
I guess we should accept use this patch but we need to get
some documentation into doc/patterns.texi. If you submit
ascii documentation I'll convert it to Texinfo.
The patch isn't in the CVS yet but I'm inclined to take it.
Dan