Re: [eia] Prussian rolls at Leipzig

[Michael Gorman <mpgorman@xxxxxxxx>]

At 11:29 AM 10/14/2004, you wrote:
> Thus spake Michael Gorman:
> >
> > Sigh.  I'm going back to the pbm dice server.  It likes me.  The nomic one
> > always rolls bad.
>
> In the interest of fairness, I decided to do some analysis on the dice
> server's random number generator to make sure that it's working properly.
>
> I took a sample of 1m rolls of a d2, converted that to bytes, and ran it
> through Ent. Here's what I got:
>
>
> [uckelman@scylla tmp]$ ./ent <c
> Entropy = 7.996935 bits per byte.
>
> Optimum compression would reduce the size
> of this 62501 byte file by 0 percent.
>
> Chi square distribution for 62501 samples is 267.50, and randomly
> would exceed this value 50.00 percent of the times.
>
> Arithmetic mean value of data bytes is 127.1886 (127.5 = random).
> Monte Carlo value for Pi is 3.166282642 (error 0.79 percent).
> Serial correlation coefficient is -0.008664 (totally uncorrelated = 0.0).
>
>
> Conclusion: These are very close to the values you'd get from a large, true
> random sample, say from a radioactive source. Recall that the reason I set up
> my own dice roller was due to the excessive latency of the one at pbm.com; I
> doubt that that's changed, so I'd recommend sticking with mine. It's up to
> you, though.

The problem with statistics is that when you only have a sample size of 
say, three rolls, it's really easy to deviate from the mean.  I'll 
naturally choose to ignore when the dice are nice to me to preserve my 
right to grump when they are not.

Mike

_______________________________________________
eia mailing list
eia@xxxxxxxxx
http://lists.ellipsis.cx/mailman/listinfo/eia