[Bug-gnubg] Re: Strange FIBS ratings

[kvandoel]

On Fri, 5 Sep 2003, Albert Silver wrote:

> I analyzed two matches in the past few days where coincidentally I made
> some huge blunders and where my error rate came out as 22+, thereby
> classifying me as a Casual Player. The problem is that the FIBS ratings
> given are VERY different, and seem a bit generous on top of it. I can't
> copy the statistics to the clipboard anymore (hint! hint!) so I'll copy
> it from the exported TXT:

> Match 1:

> Overall statistics

> Error rate (per decision)       -51.28 ( -0.542%)    -20.25 ( -0.253%)
> Error based abs. FIBS rating    1465.2               1704.0
> Chequerplay errors rating loss   353.5                344.7
> Cube errors rating loss          231.4                  1.3

> And now match 2:

> Error rate (per decision)       -13.86 ( -0.036%)    -21.97 ( -0.075%)
> Error based abs. FIBS rating    1936.0               1866.3
> Chequerplay errors rating loss    95.9                131.8
> Cube errors rating loss           18.1                 51.9

> How can I be rated 1866 with the enormous error rate in match 2?? And
> why is it so much better than the first match with a huge error rate as
> well?

The reason is that in the 2nd match your errors consisted mainly of cube
errors, which  have a  much smaller effect  on rating, which  depends on
move errors and  cube errors (not on the total  errors). The "Error rate
(per decision)" is a number which  is not well correlated with rating if
you make large cube errors.

The textual  description should be coupled  to rating and  not to "Error
rate (per decision)", I'm not sure if that has been done yet.

See

http://www.cs.ubc.ca/~kvdoel/tmp/ratings/

for the gory details.

Kees