Re: [Bug-gnubg] Results from the bot shootout

[Joseph Heled]

Even if gnubg wins a match only 49.5%, in a set of 1000 matches there
is more than 5% chance that gnubg wins 519 of them. that what 95% (one
sided) confidence interval means.

-Joseph

On 7/5/07, Achim Mueller <address@hidden> wrote:
> * Joseph Heled <address@hidden> [070705 11:14]:
> > With gnubg/Jelly we have p = .593 (estimated win percentage) and N =
> > 1000. The 95%  confidence interval is [.562,.623], so we can be sure
> > gnubg is better.
> >
> > gnubg/bgb p=.519 and the interval is [.487,.55] so you can't say gnubg
> > is better than any of the others.
>
> I understand your calculation, but what is wrong with mine? I took the
> binominal distribution and my numbers seem to be correct, aren't they?
>
> sqrt(519*481/1000) = sqrt(249,639) = 15.799
>
> actual result(+38)/sigma(15.799) = 2.4 (oops, in my first calculation I
> took +36, damn).
>
> 2.4 ---> 99.18% confidence that the same set of 1000 25p matches will
> produce a win for gnubg. Ok, I miscalculated it, but this is even more
> in favour of gnubg.
>
> Ciao
>
> Achim
>
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg