Find sum of matrices, of different sizes, in a cell

I have n matrices with different sizes in a cell and I need to find the cumulative sum, without using a for loop.

With a for loop, here is my approach

    % C = cell containing n matrices
    % size(C) = n, 1

    total = 0;
    for i = 1:n
        total = total + sum(sum(C{i}))
    end

[Jan]

Is this a homework question?

[zhirzh]

No. To be precise, I am taking part in a Coursera course - Intro to ML, by Andrew Ng. In one part, I need to find sum of N 2d matrices. for-loop was the obvious solution. I just want to find out some other solution that does not require loops.

Answer 1

[Guillaume]

It's arguable that cellfun is not a loop, but here is one way to do it

C{1}=magic(3);
C{2}=magic(5);
total = cumsum(cellfun(@(c) sum(c(:)), C)) %for cumulative sum
%or
total = sum(cellfun(@(c) sum(c(:)), C)) %equivalent to your example

Answer 2

[dpb]

s=sum(cellfun(@sum,C));

ADDENDUM For more than 1D cells, sum returns sums by column instead of single total sum. Use an anonymous function instead of native summation--

s=sum(cellfun(@(x) sum(x(:),C));

Here, use the Matlab idiom X(:) to return the vector/matrix/array X as a 1D column vector.

[dpb]

Oh, noticed you said "cumulative" when read Jan's comment -- just replace sum w/ cumsum

cs=cumsum(cellfun(@sum,C));

[zhirzh]

It didn't work.

 C=cell(2,1);
 C{1}=magic(3);
 C{2}=magic(5);
 cumsum(cellfun(@sum,C));
 error: cellfun: all values must be scalars when UniformOutput = true
 error: evaluating argument list element number 1

[Guillaume]

setting 'UniformOutput' to false is not going to help, it's just going to output a cell array that won't agree with cumsum.

[dpb]

Ah--the first was for 1D arrays (vectors)...see update for higher-dimensions.