Re: [Bug-gnubg] Fuel for the fire

[Joern Thyssen]

On Wed, Sep 17, 2003 at 09:41:21AM +1200, Joseph Heled wrote
> 
> Might be a silly idea, 

Hmm, this could be an excellent idea.

> but perhaps what we need is an equivalent to 
> "confidence interval". I

The error rate is calculated as the average error loss per "move". It's
trivial to calculate the standard deviation and 95% confidence interval.

Does this make sense?

For example,

Error rate per decision: -0.012 (+/- 0.003)

The conf. interval is -0.009 to -0.015, so the player will be rated
Intermediate to Advanced. Will this confuse users?

In Albert's example with just one cube decision the 95% CI is infinite,
thus we cannot assign a textual rating. With just two cube decisions the
interval will be large, so you may get textual ratings like "Awful to
World class".

The confidence interval can even be be applied to the FIBS rating
estimate (possibly combined with the CI Kees' calculated for his
formula).

Jørn

pgptndmnXlQ3l.pgp
Description: PGP signature