[Bug-gnubg] Re: ratings formula, checker play vs. cube

[kvandoel]

On Thu, 11 Sep 2003 14:05:06 +0200, Robert-Jan Veldhuizen wrote:

>>No  it doesn't, as  the cube  error is  multiplied by  an experimentally
>>determined factor which takes this into account.

>>So R = a*player_err + b* cube_err.

>>The effect you are worrying about is taken care of by the factor b which
>>is much smaller than a.

>Since you're using  a constant factor b,  it is only taken care  of in a
>very general way.

>For individual  games and matches,  where the factor (all  unforced cube
>decisions/close or acutal  cube decisions) can vary by  a huge amount, I
>feel pretty  confident it causes a  lot of noise and  therefore often an
>inaccurate rating estimate, even if it's good on average.

You are assuming cubeErr/totalCubeDecisions  is a better measure of cube
skill  than cubeErr/totalActualCubeDecisions. I  belive the  opposite to
hold, so what you are proposing would introduce moe noise.

>It is also  inconsistent, because checker play errors  DO get divided by
>all unforced moves.

So what?   If anything I feel  that the checker play  total error should
maybe also be  divided not just by non-forced  moves, but by non-obvious
moves. THis is just harder to quantify.


Kees