MATLAB Answers

Kobi
0

trying to minimize the memory of a matrix with sparse

Asked by Kobi
on 13 Feb 2014
Latest activity Edited by Matt J
on 13 Feb 2014
the meaning of the sparse function is to reduce memory usage by ignoring the zero elements inside a matrix, for some reason when i use this function the memory increase what am i doing wrong?
>> whos('hologram_r1_r2')
Name Size Bytes Class Attributes
hologram_r1_r2 376x376 2262016 double complex
>> hologram_r1_r2(abs(hologram_r1_r2)<min(min(abs(hologram_r1_r2))))=0;
>> whos('hologram_r1_r2')
Name Size Bytes Class Attributes
hologram_r1_r2 376x376 2262016 double complex
>> hologram_r1_r2=sparse(hologram_r1_r2);
>> whos('hologram_r1_r2')
Name Size Bytes Class Attributes
hologram_r1_r2 376x376 3396040 double sparse, complex

  0 Comments

Sign in to comment.

2 Answers

Answer by Iain
on 13 Feb 2014
Edited by Iain
on 13 Feb 2014

Sparse only saves memory if your array is dominated by 0. That is exactly zero, and not small numbers.
The memory used by sparse is: 4/8/16 bytes for the value of each nonzero element, plus some overhead to store the location of each nonzero element, so a full "sparse" array can take many times the memory of a genuine double array.

  0 Comments

Sign in to comment.


Answer by Matt J
on 13 Feb 2014
Edited by Matt J
on 13 Feb 2014

If the number of zero elements isn't at least 2/3 of the matrix (or 1/2 for a complex matrix), you won't benefit in terms of memory. Each non-zero element consumes 3 times as much memory as an ordinary matrix due to the need to store (i,j) coordinate data as well.

  1 Comment

Wide sparse matrices also tend to consume more than tall ones. Here's my favorite example of an all-zero sparse matrix, which nevertheless, consumes 80 MB of memory
>> At=sparse(1,1e7); >> At=sparse(1,1e7); whos At
Name Size Bytes Class Attributes
At 1x10000000 80000024 double sparse

Sign in to comment.