how to do multiplication for 3d array

I want to do a multiplication procedure at the 12th line (Eig_faces = C*sorted_eig_mat;) for the below code:

B = cat(3, mydata{:});
meanface = mean(B,3);
C = B - repmat(meanface, [1 1 6]);                      %shifted_image(centering) 
D = cov(B(1:end,:));                                    %cov matrix
[eig_mat, eig_val] = eig(D(:,:,1))  
eig_val_vec = diag(eig_val); 
[sorted_eig_val, eig_indices] = sort(eig_val_vec, 'descend');
sorted_eig_mat = zeros(6);
  for i=1:6
    sorted_eig_mat(:, i) = eig_mat(:,eig_indices(i));   %sorted eigen matrix
  end
Eig_faces = C*sorted_eig_mat;                           %find out eigen faces
size_Eig_faces = size(Eig_faces);

The problem is C is a vectorized value that i got based on earlier concatenate process, therefore the multiplication can't be done since the dimension is different. I got an error: error using ==> mtimes.Input arguments must be 2-D. Anyone can help me? Thank you.

[the cyclist]

What is the size of mydata{:}?

FYI, adding line numbers makes your code harder to work with. We to strip them out again to work with it.

[terrey]

i just try with a small one first, the size of mydata{:} is [2 3]

Answer 1

[Walter Roberson]

If you happened to be working with a cpuarray you could use pagefun . But since you are not, you can use arrayfun

results_cell = arrayfun( @(pane) C(:,:,pane) * sorted_eig_mat, 1:size(C,3), 'Uniform', 0);
results_array = cat(3, results_cell{:});

[terrey]

I still got an error sir.

Error using ==> mtimes. Inner matrix dimensions must agree.

Error in ==> @(pane)C(:,:,pane)*sorted_eig_mat

[Walter Roberson]

what is size(C) showing up as?

[terrey]

The size is [9,1,6] .

[Walter Roberson]

results_cell = arrayfun( @(row) permute( squeeze(C(row,:,:)) * sorted_eig_mat, [3 1 2]) , 1:size(C,1), 'Uniform', 0);
results_array = cat(1, results_cell{:});

This should result in a 9 x 1 x 6 reflecting the 9 matrices of 1 x 6 that result from multiplying the 1 x 6 by 6 x 6

[terrey]

i think i need 6 matrices of 9 x 1 because i concatenate along the third dimension sir.

[Walter Roberson]

You have

sorted_eig_mat = zeros(6);

and then you fill in columns of sorted_eig_mat without changing the size of sorted_eig_mat. So your sorted_eig_mat is 6 x 6. In order to be able to do a matrix multiplication of A*B with B being 6 x 6, your A would have to be something by 6 and the result would be (the same) something by 6. You cannot get out a 9 x 1 matrix by multiplying anything by a 6 x 6.

I suggest that for now you do not try to be fancy, and instead just code a plain "for" loop to do the multiplications. That might involve using squeeze() to eliminate an internal singular dimension after you extract a particular sub-matrix from C.