[computer-go] Re: ZG1-trunMCx-50k

[Łukasz Lew]

On 9/26/06, Don Dailey <drd at mit.edu> wrote:
> On Tue, 2006-09-26 at 01:05 +0200, Łukasz Lew wrote:
> > What do You think about advising all CGOS users to use one account for
> > one version of bot.
> >
> > And allow to decrease K indefinitely with a speed chosen to guarantee
> > convergence.
> >
> > And to increase K after each reconnect and after version/name change
> > of bot.
> >
> > Best Regards,
> > Lukasz
>
> I long ago suggested a program naming convention which is to have a base
> program name followed by a hyphen then version number.   Many people
> seem to be using this system, but it's only advisory.
>
> So you might have Foo-1.0  Foo-1.1 etc.
>
> Others I think prefer to use the same name, and update only when there
> is a major change.   But it's difficult to really know for sure.
>
> I would not consider decreasing K indefinitely but I might consider
> lowering it even more and slowing down the rate of decrease.
>

This has probably the best effort/effect ratio :)
I feel that K=3 is low enough, but bots get too fast there,
I believe that K value should decrease more slowly in the beginning
while bot is still getting
to the neighbour hood of its rating.

> The rate of decrease, by the way is not fixed, or ad-hoc.  It is based
> on the rating of the opponent.  If you play someone who has no
> experience,  your K-factor doesn't budge because from your point of
> view, nothing significant happened that gives you more confidence in
> your rating (and neither does your rating change.)
>
> It was never my intent to obsess over the rating system.  There is no
> doubt some minor tweaking possible, but there is no significant overhaul
> that will make a noticeable different, other than simply deciding what
> trade-off between stability and responsiveness we want to accept.

I agree that we are close to spent to much time on this topic. :)

Lukasz

>
>
> - Don
>
>
>