Re: [Bug-gnubg] Is GNU's luck-measure valuable?

[Christian Anthon]

Measuring luck and skill in backgammon is no easy task. It will at
best give you pointers in the right direction. The reason is that the
values are based on the programs perception of the game, which may be
more or less accurate, and the values might depend on your opponents
skill. It is, however, my experience that the values obtained are
quite good on average over a large number of games.

Christian.

On Thu, Aug 13, 2009 at 12:35 PM, Adi Kadmon<address@hidden> wrote:
> Hi all,
>
> Some data I went over today made me doubt whether GNU's luck-measure is
> valuable. A valuable measure should still leave ample room to the component
> of skill, and therfore be only party correlated with the actual results of
> games.
>
> However, when I recorded and went over exaustive data arranged in a table,
> of 29 analysed money-games with the same opponent, I came by this finding:
> in all the games, yes, in 29 out of 29 games, the actual result and the
> relative luck-measures for the two parties went in the same direction!!
>
> I should add: (1) This was always the case no matter if the difference
> between his luck and mine was big or small;  (2) his computed total error
> rates added through all the games were 50% higher than mine in checker play
> (analysed by 2-ply Supremo), also 50% higher in cube decisions (analysed by
> 2-ply World Class) and 50% higher overall.
>
> Then what about skill in backgammon? :)  Should the luck-measures be
> modified, or could anybody give me satisfactory explanations for these
> finding, corroborate or disprove them?
>
> Thanks a lot,
>
> -- Adi
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>