在Ruby中生成高斯(正态分布)随机数的代码

What is some code to generate normally distributed random numbers in ruby?

在ruby中生成正态分布随机数的代码是什么?

(Note: I answered my own question, but I'll wait a few days before accepting to see if anyone has a better answer.)

(注:我已经回答了自己的问题，但我要等上几天，才能接受是否有更好的答案。)

EDIT:

编辑:

Searching for this, I looked at all pages on SO resulting from the two searches:

搜索这个，我查看了所有的页面，所以从这两个搜索结果:

+"normal distribution" ruby

ruby +“正态分布”

and

和

+gaussian +random ruby

+高斯随机ruby

4 个解决方案

#1

Python's random.gauss() and Boost's normal_distribution both use the Box-Muller transform, so that should be good enough for Ruby too.

Python的随机性.gauss()和Boost的normal_distribution都使用Box-Muller转换，因此对Ruby来说也应该足够好。

def gaussian(mean, stddev, rand)
  theta = 2 * Math::PI * rand.call
  rho = Math.sqrt(-2 * Math.log(1 - rand.call))
  scale = stddev * rho
  x = mean + scale * Math.cos(theta)
  y = mean + scale * Math.sin(theta)
  return x, y
end

The method can be wrapped up in a class that returns the samples one by one.

方法可以封装在一个类中，该类逐个返回示例。

class RandomGaussian
  def initialize(mean, stddev, rand_helper = lambda { Kernel.rand })
    @rand_helper = rand_helper
    @mean = mean
    @stddev = stddev
    @valid = false
    @next = 0
  end

  def rand
    if @valid then
      @valid = false
      return @next
    else
      @valid = true
      x, y = self.class.gaussian(@mean, @stddev, @rand_helper)
      @next = y
      return x
    end
  end

  private
  def self.gaussian(mean, stddev, rand)
    theta = 2 * Math::PI * rand.call
    rho = Math.sqrt(-2 * Math.log(1 - rand.call))
    scale = stddev * rho
    x = mean + scale * Math.cos(theta)
    y = mean + scale * Math.sin(theta)
    return x, y
  end
end

在Ruby中生成高斯(正态分布)随机数的代码 (CC0)

(CC0)

To the extent possible under law, antonakos has waived all copyright and related or neighboring rights to the RandomGaussian Ruby class. This work is published from: Denmark.

在法律允许的范围内，antonakos放弃了随机性高斯Ruby类的所有版权和相关或邻近的权利。本著作出版于:丹麦。

The license statement does not mean I care about this code. On the contrary, I don't use the code, I haven't tested it, and I don't program in Ruby.

许可声明并不意味着我关心这段代码。相反，我不使用代码，也没有测试它，也没有在Ruby中编程。

#2

The original question asked for code, but the author's followup comment implied an interest in using existing libraries. I was interested in the same, and my searches turned up these two ruby gems:

最初的问题需要代码，但是作者的后续评论暗示了对使用现有库的兴趣。我也有同样的兴趣，我的搜索发现了这两个红宝石宝石:

gsl - "Ruby interface to the GNU Scientific Library" (requires you to install GSL). The calling sequence for normally distributed random numbers with mean = 0 and a given standard deviation is

gsl -“GNU科学图书馆的Ruby接口”(要求您安装gsl)。均值为0且给定标准差的正态分布随机数的调用序列为

 rng = GSL::Rng.alloc
 rng.gaussian(sd)      # a single random sample
 rng.gaussian(sd, 100) # 100 random samples

rubystats - "a port of the statistics libraries from PHPMath" (pure ruby). The calling sequence for normally distributed random numbers with a given mean and standard deviation is

rubystats——“来自PHPMath的统计库的一个端口”(纯ruby)。具有给定均值和标准差的正态分布随机数的调用序列为

 gen = Rubystats::NormalDistribution.new(mean, sd)
 gen.rng               # a single random sample
 gen.rng(100)          # 100 random samples

#3

+1 on @antonakos's answer. Here's the implementation of Box-Muller that I've been using; it's essentially identical but slightly tighter code:

+ 1 @antonakos的答案。这是我一直在使用的Box-Muller的实现;它本质上是相同的，但更紧密的代码:

class RandomGaussian
  def initialize(mean = 0.0, sd = 1.0, rng = lambda { Kernel.rand })
    @mean, @sd, @rng = mean, sd, rng
    @compute_next_pair = false
  end

  def rand
    if (@compute_next_pair = !@compute_next_pair)
      # Compute a pair of random values with normal distribution.
      # See http://en.wikipedia.org/wiki/Box-Muller_transform
      theta = 2 * Math::PI * @rng.call
      scale = @sd * Math.sqrt(-2 * Math.log(1 - @rng.call))
      @g1 = @mean + scale * Math.sin(theta)
      @g0 = @mean + scale * Math.cos(theta)
    else
      @g1
    end
  end
end

Of course, if you really cared about speed, you should implement the Ziggurat Algorithm :).

当然，如果你真的关心速度，你应该实现Ziggurat算法:)。

#4

Another option, this one using the distribution gem, written by one of the SciRuby fellows.

另一个选择，这个使用发行版gem，由一个SciRuby伙伴编写。

It is a little simpler to use, I think.

我认为它使用起来比较简单。

require 'distribution'
normal = Distribution::Normal.rng(1)
norm_distribution = 1_000.times.map {normal.call}

#1