Re: [Bug-gnubg] FIBS rating and error rates

[Joern Thyssen]

On Sat, Aug 16, 2003 at 11:13:19AM +0200, address@hidden wrote
> Hi,
> 
> I recently completed  a series of simulations in  order to determine the
> relation between normalized error rate  per decision and FIBS rating. If
> I didn't mess  up these results will allow GNUBG  to report the absolute
> rating of  a player from  an error analysis  (i.e., the feature  that hs
> been removed recently).
> 
> The results  are summarized on http://www.cs.ubc.ca/~kvdoel/tmp/ratings,
> and I'd appreciate comments,  criticism, and suggestions for refinements
> if necessary.

Have you tried finding a function that fits a(N) in the  "linear coeff
versus match length" graph?

Or should we just use the piecewise linear interpolation as drawn in the
graph?

Jørn

pgphu9U8FIeIl.pgp
Description: PGP signature