Re: [gnugo-devel] How to change the influence function?

[Stuart A. Yeates]

There is a huge body over work on this problem, which is called
"overfitting" in the general case.

The solution is to judge quality based on games as independent as
possible. Ideally against human players even.

cheers
stuart


On Fri, Jul 18, 2008 at 6:38 PM, Teo Swee Ann <address@hidden> wrote:
> I would like to add that, winning is not transitive: If A>B>C, it doesn't
> always mean that A>C, sometimes; at least in Go, where ">" means wins.
>
> Perhaps what we need is to show that A > (all X, where X <= B), then we can
> conclude that A is "transitively stronger" than B, with reference to a
> specific set of X's.
>
> Perhaps if we define this set of reference opponents, we can then
> incrementally improve the strength of the Go programs, slowly but
> definitely.
>
> Swee-Ann
>
> -----Original Message-----
> From: address@hidden
> [mailto:address@hidden On Behalf Of
> Darren Cook
> Sent: Friday, July 18, 2008 2:10 PM
> To: GNU Go development
> Subject: Re: [gnugo-devel] How to change the influence function?
>
> Hi,
> I'll just answer the non-gnugo-specific question :-).
>
>> I believe it can be said the new influence function is better if the
>> new GNU Go wins more games, in average, than the old one.
>
> Yes.
>
>> In other words, this new influence
>> function should be compared from self-play (old influence function
>> vs. new influence fn). Does this sound reasonable?
>
> No, tuning against just one opponent is risky.
> Instead set up a tournament of 3-4 programs with different styles. (Or
> the simpler way to do that is to put it on cgos:
> http://senseis.xmp.net/?ComputerGoServer )
>
> Darren
>
>
> --
> Darren Cook, Software Researcher/Developer
> http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic
>                        open source dictionary/semantic network)
> http://dcook.org/work/ (About me and my work)
> http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...)
>
>
> _______________________________________________
> gnugo-devel mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gnugo-devel
>
>
>
> _______________________________________________
> gnugo-devel mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gnugo-devel
>