Summing identical submatricies to get a smaller final matrix

Hi everyone,

Here is my problem, I have a 2500x2500 matrix, this can be thought of as (50x50)x(50x50). I would like to sum each 50x50 sub-matrix and output a smaller 50x50 matrix where each point in the matrix represents the sum of each larger 50x50 matrix.

I have tried using blockproc:

if true
b_cx=JXrPrimex.*InvrMPC;
fun = @(block_struct) ...
  sum(b_cx)
Bcx=blockproc(b_cx,[50 50], fun);
  % code
end

However, when I run 'whos Bcx' I get a value that isn't a 50x50 matrix.

I also experimented with mat2cell but thought this seemed inefficient as after breaking each matrix down into 50x50 submatricies, I would then have to sum them all and then reconstruct a matrix of these sums. This is because my final answers needs to be in matrix not cell form.

Hope I was clear, if I wasn't comment and I will do my best to explain.

Thanks, John.

Answer 1

[Guillaume]

Firstly, sum applied to a matrix return a row vector, not a scalar. Secondly, your fun function does not work on a block but on the whole matrix. Your fun function should be:

fun = @(block_struct) sum(block_struct.data(:)); %or sum(sum(block_struct.data))

As you said mat2cell would also work, the code is very simple and in my testing a lot faster than blockproc:

b_cx=JXrPrimex.*InvrMPC;
split_bcx = mat2cell(b_cx, ones(1, size(b_cx, 1)/50)*50, ones(1, size(b_cx, 2)/50)*50);
Bcx = cellfun(@(c) sum(c(:)), split_bcx);

Answer 2

[Matt J]

Using SEPBLOCKFUN

A=reshape(yourMatrix,50,50,50,50);

B=sepblockfun(A,[50,50,1,1],'sum');

result = reshape(B,[50,50]);