RE: [Bug-gnubg] Gnubg's ratings and checker versus cube errors.

[Misja Alma]

Thanks for the links!
I read Kees' article, and one thing that struck me was the following:
The difference in rating between 2 players is calculated from the total sum
of their errors. That could be all their errors or a weighted sum of cube-
and checker play errors.
Is that the normalized sum or just the total sum of the errors? I read in
one post the following formula:
errTot = moveErr/X + C*cubeErr/X, with
> X=(unforcedMoves+unforcedCubedecisions)

Now this total can serve as a basis to calculate a rating, and I read that
Kees' formula does a pretty good job at this.
But just adding up the errors that are made without normalizing them, does
not calculate the winning change of player A against player B. If you just
add everything up you can end up with a winning chance of more than 100%.
I think the right way of adding up errors is this:
Pw2 = Pw1*(1-Err)
Where Err is the percentage lost by the error and Pw1 the estimated winning
chance of the player before the error.
When a second error is made Pw3 = Pw2*(1-Err2) etc.
I think this is the only way of calculating a winning chance of player A
versus player B which lies between 0 and 100%, or am I wrong?

Misja

-----Oorspronkelijk bericht-----
Van: address@hidden
[mailto:address@hidden Holger
Verzonden: Wednesday, December 03, 2003 8:19 PM
Aan: Misja Alma; address@hidden; address@hidden
Onderwerp: RE: [Bug-gnubg] Gnubg's ratings and checker versus cube
errors.


At 17:55 03.12.2003, Misja Alma wrote:
>I don't understand why there should be different ratings for cube and
>checker play errors? Is it perhaps not a good idea to base the rating on
>lost game/ match winning chances regardless if it was a cube error or a
>checker play error?

See my other reply in this thread and Kees' tests.

Regards,

         Holger



_______________________________________________
Bug-gnubg mailing list
address@hidden
http://mail.gnu.org/mailman/listinfo/bug-gnubg