蒙特卡罗方法的矢量化近似pi

时间:2014-02-22 14:31:02

标签: matlab

成功之后,我编写了以下代码,基于for循环使用蒙特卡罗方法近似数字pi:

function piapprox = calcPiMC(n)
count = 0;      % count variable, start value set to zero
for i=1:n;      % for loop initialized at one
    x=rand;     % rand variable within [0,1]
    y=rand;     

    if (x^2+y^2 <=1)    % if clause, determines if pair is within the first quadrant of the unit circle
    count = count +1;   % if yes, it increases the count by one
    end

end

piapprox = 4*(count/n);

end

这种方法很棒但是对于大n的大值开始挣扎很多。作为我自己的一项任务,我想尝试对以下代码进行矢量化,不使用使用任何循环等。在我脑海中,我可以想到一些对我来说有意义的伪代码,但到目前为止我无法完成它。这是我的方法:

function piapprox = calcPiMCVec(n)

count = zeros(size(n));    % creating a count vector filled with zeros
x = rand(size(n));         % a random vector of length n
y = rand(size(n));         % another random vector of length n

if (x.*x + y.*y <=  ones(size(n)))   % element wise multiplication (!!!)
    count = count+1;                 % definitely wrong but clueless here
end

piapprox=4*(count./n); 
end 

我希望在阅读这个伪代码时我的想法看起来很清楚,但我会对它们发表评论。

  • 我开始创建一个零的计数向量,其长度由向量n
  • 的大小决定
  • 接下来,我对随机条目xy
  • 做了同样的事情
  • if子句应该确定哪些条目小于1,与填充了相同长度的向量相比,但我很确定这段代码是错误的。
  • 最后我想将if子句为true的向量数存储到向量中,这段代码也错了,我想我必须在这里使用cumsum函数,但是结果出错了

1 个答案:

答案 0 :(得分:4)

这是一个矢量化解决方案,其中包含一系列评论,部分参考了您的尝试。

function piapprox = calcPiM(n)
%#CALCPIM calculates pi by Monte-Carlo simulation
%# start a function with at least a H1-line, even better: explain fcn, input, and output

%# start with some input checking
if nargin < 1 || isempty(n) || ~isscalar(n)
   error('please supply a scalar n to calcPiM')
end

%# create n pairs of x/y coordinates, i.e. a n-by-2 array
%# rand(size(n)) is a single random variable, since size(n) is [1 1]
xy = rand(n,2); 

%# first, do element-wise squaring of array, then sum x and y (sum(xy.^2,2))
%# second, count all the radii smaller/equal 1
count = sum( sum( xy.^2, 2) <= 1);

%# no need for array-divide here, since both count and n are scalar
piapprox = 4*count/n;