[Bug-gnubg] FIBS ratings

[Albert Silver]

> The problem is that the two rating estimates will be inconsistent,
> e.g., the two absolute ratings could be 1800 and 1900, but the
> difference is estimated to be 200.... Note that it's not possible to
> "syncronise" the two, since they're based on different numbers (error
> rate and luck adjusted result).
> 
> A long time ago gnubg also produced two rating estimates which were
> inconsistent, and I got bombed with mails asking why there was a
> difference, so I'm not opening that can of worms again!
> 
> I'm happy to reinstate a "gnubg rating" if someone produces some
> material that relates normalised error rates per move and match length
> with some estimate of winning the match.
> 
> Jørn

Although anyone can keep a record of their FIBS matches and send them,
this might be slow to collect enough material. Why not ask Heled for
his log of GNU 0-ply games. This way GNU 2-ply would be able to
analyze the errors of its slightly more error prone sibling and give
some info on its relative rating. Watch out for the rating
manipulators such as Nihilist though, as that will skew the results.

One suggestion: the issue is estimating a FIBS rating or FIBS rating
performance in the absence of information, but how about having it take
into account the ratings if they are provided by the user in the match
information area?

                                     Albert