RE: [Bug-gnubg] Re: Strange FIBS ratings

[kvandoel]

> > Let's just collect some facts instead.
>
> Indeed, I did. I took the last 30 matches (all of them AFAIK), many of
> which I had saved with analyses still, and others that I had to redo. I
> used 0-ply checker and 2-ply cube (I don't trust 0-ply cube and twice
> even debunked the 2-ply with a rollout) - rerunning all errors through
> 2-ply - so as to not take excessively long on this. The results were a
> combination of exactly what I said (completely off), and exactly what
> you said (very accurate), so now that I have the numbers I'm literally
> forced to wonder about bad luck. Perhaps you'll have a different
> explanation. Attached is the Excel file, but here is what it says about
> the 30 matches (123 games) in a nutshell:
>
> My rating is of course ~1700 and my average FIBS rating according to GNU
> is 1871.1 which supports my comment it would seem.

It depends.  What is the  confidence interval? It  could well be  200 in
which case your data doesn't show anything.

> Still, there is the other data on my opponents. Their average FIBS
> rating was 1732 and their average attributed rating was 1762.1, which
> supports your simulations.

Same comment.

> I scored 50%, which might also make one wonder why my rating is not a
> bit higher. Just to be clear, I save all my matches in the same place,
> and have not deleted any, so to my knowledge this is all of them. The
> last time I made a cleanup was about 3-4 weeks ago (according to the
> dates of the files).

Would you mind sending me the 50 matches in .mat format so I can analyze
them with my own methods and compare results? I use 0-ply everywhere and
a comparison would help to evaluate how reliable that is too.

I didn't see any excel attachement.

Thanks,
Kees