Substract nd array slices

Hi,

I have a n-dimensional array which may contain singleton dimensions and I would like to substract the second page of dimension i from the first page of dimension i. My current solution looks something like this:

function A = substract_page( A, dim )

  orig_size = size(A);    % store the original shape

  A = shiftdim(A, dim);   % shift the dimension to be substracted to the end
  shifted_size = size(A); % aliter: shifted_size = circshift orig_size(dim); ?
  A = reshape(A, [], shifted_size(end));  % collapse first dimensions for easier substraction
  A = A(:, 2) - A(:, 1);
  shifted_size(end) = shifted_size(end) - 1;  % substraction leads to dimension loss
  A = reshape(A, shifted_size);               % restore shape
  lost_dimensions = length(orig_size) - ndims(A);
  A = shiftdim(A, -lost_dimensions);         % make sure that  singular dimensions  in the end don't get lost
  A = shiftdim(A, length(orig_size) - dim + lost_dimensions);    % restore order

end

But it seems way to complicated and I feel there is a much easier and more elegant solution.

Answer 1

[Jan]

Does your code perform this:

dim = 3;
A   = rand(2,3,4,5);
A   = A(:, :, 2, :) - A(:, :, 1, :)

? Then:

function A = substract_page(A, dim)
C1      = repmat({':'}, 1, ndims(A));
C2      = C1;
C1{dim} = 1;
C2{dim} = 2;
A       = A(C2{:}) - A(C1{:});
end

[Orangelynx]

good answer, but `diff` works even better (at least i haven't found a scenario where it doesn't perform as expected yet).

[Jan]

@Orangelynx: When the dimension to operate on has the length 2, diff is perfect. My code would be useful only, if the input has e.g. the length 4 as in my example, and only the difference between the first 2 slices should be calculated ignoring the rest.

[Orangelynx]

good point, I guess I didn't specify that clearly in my original question.

Answer 2

[Matt J]

Not sure, but maybe you are looking for

A=diff(A,1,dim);

[Orangelynx]

wow.. I knew about diff, but for some reason my head told me it's not what I was looking for, eventhough with the dim parameter, it's exactly what I wanted... thanks.