Re: [computer-go] future KGS Computer Go Tournaments - two sections?

["Jim O'Flaherty, Jr." <jim_oflaherty_jr@xxxxxxxxx>]

Christoph,

How many games would you need to play A versus B in order to come to an acceptable (high?) degree
of confidence about the percentage of time A wins over B?

This is why I posted earlier about the degree of randomness present in a computer Go competition. 
If the goal is to find the *consistent* ranking of a particular program/configuration as compared
to another program/configuration, unless randomness is accounted for, it seems the ranking is more
about which got lucky what percentage of the time as opposed to which program has superior
performance.

As to the value of other rankings (i.e. teams, amount of time spent, etc.), those just seem that
much more "ambiguous" as measurements where any reasonable objective measure could be defined.


Jim


--- Christoph Birk <birk@xxxxxxxxxxxxxxxxx> wrote:
> > I think the goal is that you don't want two (or more) basically 
> > indistinguishable copies of a program playing in the same tournament.  
> 
> Another simple way to define "sufficiently different" is to 
> compare the strength of the program. If 'B' wins agains 'A'
> more than eg. 60% of the time, I would consider 'B' different
> from 'A', whether it's derived from 'A' or not.
> 
> Christoph
>  
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
> 
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/