From: Joel VanderWerf <vjoel@...>
Date: 2009-01-27T05:08:57+09:00
Subject: Re: Separate random number generators?

Bart Braem wrote:
> Do not use ISAAC, running this simple test:
> 
>           rng = Crypt::ISAAC.new
>           res = Hash.new()
>           1.upto(100000) do |i|
>             random = rng.rand
>             larger = (random * 10).to_i
>             res[larger] ||= 0
>             res[larger] += 1
>           end
>           1.upto(10) do |i|
>             puts "hits for #{i}: #{res[i]}"
>           end
> 
> Shows a large bias towards numbers below 0.4, unless I have an error
> in this simple script of course.
> rb-gsl gives problems with installation, I am going to try the Drb
> solution.

I use my own wrapper around Bob Jenkins' ISAAC library. Based on his 
comments on ruby-talk a few years ago, I chose to use the 256 rather 
than 16 entry state vectors (it turns out not to affect performance). 
It's served well for simulation purposes.

Your code, modified slightly, gives the following results:

hits for 0: 10213
hits for 1: 9830
hits for 2: 9902
hits for 3: 9928
hits for 4: 9951
hits for 5: 10111
hits for 6: 10012
hits for 7: 9940
hits for 8: 9896
hits for 9: 10217

Here's the example:

require 'isaac'

rng = ISAAC.new
#rng.srand([...]) seed with up to 256 entries (optional)
res = Hash.new()
1.upto(100000) do |i| # 100000
   random = rng.rand
   larger = (random * 10).to_i
   res[larger] ||= 0
   res[larger] += 1
end
0.upto(9) do |i|
   puts "hits for #{i}: #{res[i]}"
end


The extension is at:

http://redshift.sourceforge.net/isaac-0.1/

-- 
       vjoel : Joel VanderWerf : path berkeley edu : 510 665 3407