From: Martin Jansson <martialis@...>
Date: 2005-09-11T16:41:32+09:00
Subject: Re: Better Random benchmark - need help with profiling

On Sat, 10 Sep 2005 21:34:44 +0200, Brian Schr�der <ruby.brian@gmail.com>  
wrote:

> class RandomNumberGenerator
>   def initialize(seed = 42)
>     @seed = seed
>   end
>  def random(max)
>     (max * (@seed = (@seed * 3877 + 29573) % 139968)).quo(139968)
>   end
> end
> rng = RandomNumberGenerator.new
> (ARGV[0].to_i-1).times do
>   rng.random(100)
> end
> puts "%.9f" % rng.random(100)


> bschroed@black:~/svn/projekte/random$ ruby benchmark-random.rb
                                     user     system      total        real
> Original                        0.020000   0.050000   9.620000 (   
> 9.583356)
> Rubiesque                       0.010000   0.060000   9.820000 (   
> 9.792614)
>As you see, it is quite cheap to write prettier code. And the modified
> version even keeps the state in a class-variable, so you can have
> multiple rng's with differen states around.

Shroeder! What version of Ruby do you use?

I did something similar in one of my many versions. And it is the oo thing  
to do.
(I haven't actually benchmarked puts vs printf, I will do that ;)

It's not faster on my system. That may be proof that profiling is a bad  
thing and that we all should prove our code mathematically (shudder).

martinjansson$ time ruby "random_shroeder.rb" 900000
75.544410151

real    0m10.296s
user    0m9.760s
sys     0m0.060s

martinjansson$ time ruby "random.rb" 900000
75.544410151

real    0m9.984s
user    0m9.230s
sys     0m0.020s

martinjansson$ ruby "random_benchmark.rb" 200 5000
                                     user     system      total        real
Ugly                            0.100000   0.380000  24.770000 ( 26.368260)
Rubiesque                       0.130000   0.370000  25.580000 ( 27.071112)

martinjansson$ ruby "random_benchmark.rb" 10 900000
                                     user     system      total        real
Ugly                            0.000000   0.010000  85.770000 ( 98.821162)
Rubiesque                       0.000000   0.020000  99.810000 (105.068127)


Here is the benchmark code:

#random_benchmark.rb

require 'benchmark'
i = ARGV[0].to_i
j = ARGV[1].to_i

Benchmark.bm(30) do | b |
   [["random.rb", "Ugly"],
   ["random_shroeder.rb", "Rubiesque"]].each do | file, title |
     b.report(title) do
     i.times do system "ruby #{file} #{j} > /dev/null" end
     end
   end
end

I have a lot of white noise on my system, so I haven't used Rubys  
benchmark thingie much. It would be more useful to me if it could do  
matched pair sampling (is that the correct English term?) and/or throw  
away extreme values.