[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] GNU Go 3.3.6

From: Evan Berggren Daniel
Subject: Re: [gnugo-devel] GNU Go 3.3.6
Date: Sun, 18 Aug 2002 23:54:28 -0400 (EDT)

On Sat, 17 Aug 2002 address@hidden wrote:

> > Evan wrote:
> > > so, any indication how much stronger it is than 3.3.5?  3.2?  is this
> > > something we traditionally check at point releases, or not really?
> >
> > No, we usually don't although the information is potentially useful.
> > The problem is that in order to measure smallish incremental changes,
> > some 500 or so games need to be played before the statistics become
> > reliable. Obviously this requires a substantial amount of
> > computations, which is the main difficulty to overcome. At one time
> > Jerome Dumonteil did this kind of testing and the (now rather old)
> > results can still be found on the pages linked from the development
> > page.
> We did a lot of testing of this type using twogtp before the 3.2
> release. We were able to say confidently that 3.2 could give 3.0
> a 3-4 stone handicap and win most of the time.

OK, I'm running 100 games between the two, no handicap (3.3.6 vs 3.2).
Currently, 9 games are done, and 3.3.6 is ahead 6-3 with larger margins
when it wins.  I'll post game records when it's done (3-4 days?).

Should I change to a handicap on the preliminary results or finish it out


Evan Daniel

reply via email to

[Prev in Thread] Current Thread [Next in Thread]