Convert each 2D matrix in a collection of matrices into a diagonal matrix

I have a collection of 2D matrices stored in a 3D matrix, and I want to keep only the elements on the main diagonal of each one. I managed to do this a shown below, how would this be done more efficiently?

A = rand(N,M,M); % my matrix is not really random, this is just an example.
for i = 1:N
    A(i,:,:) = diag(diag(squeeze(A(i,:,:))));
end

[KALYAN ACHARJYA]

May be this one

https://www.mathworks.com/matlabcentral/answers/81570-indexing-diagonals-out-of-a-3d-matrix

Answer 1

[Guillaume]

A = A .* permute(eye(size(A, 2), [3 1 2]))   %multiply each matrix by the identity matrix which keeps just the diagonal of each

Note that it's a bit odd to stack your matrices along the first dimension. You'd be better off stacking along the third dimension which would remove the need for the squeeze in your loop, and the need for the permute in the above.

I.e. you'd be better off storing A as:

permute(A, [2 3 1])  %move dim 2 into 1st dimension, dim 3 into 2nd dimension, and dim 1 into 3rd

[Evripidis Papaevripidis]

Thanks, Just what I was looking for!