[PD] Random without repeating numbers

[Mathieu Bouchard]

On Thu, 28 Nov 2002, Frank Barknecht wrote:
> David N G McCallum hat gesagt: // David N G McCallum wrote:
> > 	Maybe I'm missing something here, or it's already been said, but
> > couldn't you just set up a really simple patch to compare the output of
> > [random] with its previous output. If new==old then bang the random
> > again, if new!=old then pass it through.
> That's what I thought at first glance, too, but as randomness goes, it
> is possible, that you get a lot of equal numbers after another and
> that would stop your flow of numbers.

If you can afford to do N retries without desyncing your patch, and the
probability of a good number is 80%, then the probability of a desync will
be 20%^N. For example, for N=10, that's 1 out of about 10 million. That's
if your uniform generator [random] is uniform enough -- if not, then
strange things may happen.

> Mathieu's solution is a very elegant algorithm, I didn't know before,
> but I'll remember that one now.

Thanks =)

> I think it was a swiss scientist, who tested how "random" humans
> behave. He rolled a dice, and test persons were told to guess the
> result without seeing the dice. The humans unintentionally tried to
> avoid repeating numbers. Subconsciously they must think, that "true
> randomness" excludes repeating results. Of course it doesn't.

I remember a prof asking half of the class to throw a coin 100 times and
note the results. And the other half to fake them. He was then trying to
figure out which ones were faked just by looking at the sheet. His
criterion was that real results typically include 6 times the same value
in a row (TTTTTT or HHHHHH) while faked results typically avoid those.

________________________________________________________________
Mathieu Bouchard                       http://artengine.ca/matju