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

[kvandoel]

Hi,

I recently completed another series of simulations in order to determine
the relation  between error rates  and FIBS rating. The  new simulations
resulted in a much improved formula,  using a bilinear fit to move error
and cube error separately.

The results  are summarized on http://www.cs.ubc.ca/~kvdoel/tmp/ratings,
and I'd appreciate it if anyone  could have a critical look, making sure
I didn't mess up anywhere.

Bottom line is the following formula for the rating difference:

R = a2(N)*EPM+b(N)*EPC,
where
a2(N) = 8798 + 2526/N,
and
b(N) = 863 - 519/N.


Kees