Calculate max(diff(A)) fast and memory efficient.

Question

Matthias 2014-6-3

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/132299-calculate-max-diff-a-fast-and-memory-efficient

回答： James Tursa 2014-6-3

在 MATLAB Online 中打开

Say I have a giant array A of 2 gb and I want to make something like an analog edge-detection:

[max, index] = max(diff(A))

This would require at least 4 gb of ram. There is however a more memory efficient way:

A = randi(1e5, 1,5e8);
tic;
[m, index] = max(A);
toc;
disp(index);
tic;
m = 0;
index = -1;
for i = 1:length(A)
    if A(i) > m
        m = A(i);
        index = i;
    end
end
toc;
disp(index);

which outputs:

Elapsed time is 36.869092 seconds.
       37662
Elapsed time is 45.502829 seconds.
       37662

In any programming language with a decent jit the lower part would be as fast or faster than the upper. Is there a fast way to implement this in matlab?

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Matt J 2014-6-3

编辑：Matt J 2014-6-3

在 MATLAB Online 中打开

Your example is a bit confusing because it excludes the diff operations. When run as shown, the max(A) approach gives

Elapsed time is 0.583208 seconds.

Also, the diff operation is presumably what is responsible for the extra 2GB RAM usage.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

James Tursa 2014-6-3

2
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/132299-calculate-max-diff-a-fast-and-memory-efficient#answer_139376

在 MATLAB Online 中打开

You could resort to a mex routine. E.g., here is bare bones code (no argument checking) for double input:

#include "mex.h"
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
  double d, f;
  double *pr, *qr;
  mwSize i, n, x;
  n = mxGetNumberOfElements(prhs[0]);
  qr = mxGetPr(prhs[0]);
  pr = qr + 1;
  x = 1;
  d = *pr - *qr;
  for( i=2; i<n; i++ ) {
    f = *++pr - *++qr;
    if( f > d ) {
      d = f;
      x = i;
    }
  }
  plhs[0] = mxCreateDoubleScalar(d);
  if( nlhs == 2 ) {
      plhs[1] = mxCreateDoubleScalar(x);
  }
}

This will avoid the intermediate data copy. To compile it, place the code inside a file called maxdiff.c in your working directory, make that working directory your current directory, then type this at the MATLAB prompt:

mex maxdiff.c

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

Answer 2

Sean de Wolski 2014-6-3

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/132299-calculate-max-diff-a-fast-and-memory-efficient#answer_139362

You could do a hybrid approach and loop from 1:4 (or 10 or whatever), calculate max(diff(x(of that range of x)) and then keep the biggest one at the end.