Better way to use circshift

I am trying to create array B from array A (of course in the real application the numbers are not sequential and the arrays are much biggger i.e. 5000x5000)

B =

     5     0     1     2     3     4
     4     5     0     1     2     3
     3     4     5     0     1     2
     2     3     4     5     0     1
     1     2     3     4     5     0

A =

     0     1     2     3     4     5
     0     1     2     3     4     5
     0     1     2     3     4     5
     0     1     2     3     4     5
     0     1     2     3     4     5

Here is my solution:

for i = 1:5 B(i,:) = circshift(A(i,:),[0,i]); end

Is there a better (i.e. faster) way to do this? On a more general note, I get the fealing that using a for loop to act on a array is always a bad idea.

[Image Analyst]

Wrong. Using for loops is not always a bad idea.

[Brent]

Thanks for the answers. I didn't realize I would spark a debate. Unfortunatly, my array is not sparse. However, thaks to your comments, I realized that B is just a Toeplitz matrix based on the rows in A (duh ... can't believe I didn't see that). Here is the final code that I came up with:

 r = rand(5000,1);
>> A = ones(size(r,1)-1,1)*r';
>> tic;B = toeplitz(fliplr(r'),circshift(r',[0,1]));toc
Elapsed time is 0.474317 seconds.

Not as good as Matt J., but it will do for my purpose. Thanks

Answer 1

[Matt J]

If the matrix is going to be sparse, this should be pretty fast

http://www.mathworks.com/matlabcentral/fileexchange/26292-regular-control-point-interpolation-matrix-with-boundary-conditions

   B=interpMatrix([5 4 3 2 1],1,6,1,'circ');
   B(end,:)=[];

Answer 2

[Sean de Wolski]

v = 0:5;
B = gallery('circul',v);
B = B(2:end,:);
A = repmat(v,[numel(v)-1,1]);

[Matt J]

Careful. GALLERY has had a hard time producing large sparse circulant matrices, as discussed here

http://www.mathworks.com/matlabcentral/newsreader/view_thread/314801#859946

Don't know if they fixed this in more recent MATLAB versions...

[Sean de Wolski]

tic,B = gallery('circul',1:5000);toc

Elapsed time is 0.603616 seconds.

[Matt J]

It seems to be doing better in terms of memory management, compared to past MATLAB versions, but it's still slow. Take the example from my link. In R2012a, I get these timings:

   r=[1 2 3 4 sparse(zeros(1,10000))];

   tic; A=gallery('circul',r); toc

Elapsed time is 4.887724 seconds.

   tic; A=interpMatrix(r,1,length(r),1,'circ').'; toc

Elapsed time is 0.005704 seconds.