[ILUG] how to answer the question: "how random?"

[Kenn Humborg]

> On Fri, 27 Apr 2001 16:40:15 kevin lyda wrote:
> 
> > how can i analyse a stream of numbers to see how random they are?
> > does anyone have any web pages that discuss this?  i napped too much
> > in stats - a bad course to take in the spring, even in buffalo it got
> > too hot for stats around april...
> > 
> > i'm looking to compare the numbers returned from egd to the ones
> > returned
> > from /dev/random on linux (and maybe openbsd if i'm bored).
> 
> Well, one thing you could do is graph and compare the distribution of
> each.
> 
> You should do the tests multiple times. Also if you're dividing the range
> into "n" buckets, output a graph every n*10 samples and take at least
> n*100 samples. You may want to measure the standard deviation at each
> stage, or you may not as it might not mean enough to you if you slept
> through stats.

Well, the "random" sequence 0 1 2 3 4 5 6 7 8 9 <repeat> will give
you a perfectly even distribution, but very little randomness.

Get a fourier transform of a large sample.  Perfectly random
data will have equal "power" at all "frequencies" (terms power
and frequency make more sense to us electronics guys - pure
white noise has a flat frequency spectrum).

Later,
Kenn