From: Caleb Clausen <vikkous@...>
Date: 2010-04-12T02:30:18+09:00
Subject: [ruby-core:29452] Re: [RubySpec #3128] Randomness specs

On 4/10/10, Marc-Andre Lafortune <redmine@ruby-lang.org> wrote:
> RubySpec #3128: Randomness specs
> http://redmine.ruby-lang.org/issues/show/3128
>
> Author: Marc-Andre Lafortune
> Status: Open, Priority: Normal
> Category: core, Target version: 1.9.2
>
> What should be the Ruby specs for the new Random class (and existing
> Kernel.{s}rand)?
>
> More precisely: what should one expect of any Ruby implementation?
>
> Several degrees of similarity with MRI are possible:
>
> Say r = Random.new(42) and N is an Integer
>
> 0) r.rand(N) is included in 0...N

Yes... and r.rand(N) should be an Integer.

> 1) r.rand(N) will eventually return all values in 0...N

I find this troubling, since it could imply that the random number
generator obeys the mythical law of averages. You know better, I'm
sure, but others may not. A written standard needs to be very clear
about this, and use the appropriate statistical language.

> 2) r.rand(N) will return any particular value in 0...N with a probability of
> around 1/N
> <insert any any of the known random tests and/or a minimal period>
> 3) r is a Mersenne Twister
> 4) r is MT19937
> 5) r.rand(N) generates the same particular string on all platforms

5) could be interpreted 2 ways, depending on the meaning of the
ambiguous term 'platform'.

5a) All ruby implementations must generate the same sequence of random
numbers as MRI from r.rand(N).

5b) A particular implementation must generate the same sequence of
random numbers from r.rand(N), no matter what operating system it is
used on.


Of course, 5) only applies to pseudo-random number generators. One can
easily imagine a subclass of Random which uses a truly random number
generator.

Testing (pseudo)random number generators is difficult, due to their
(pseudo)nondeterministic nature.