gather data from different sections of each column

Consider the code:

 totmods    = 10000;
 numselect = 1000;
 A = rand(100,totmods);        % numbers
 B = randi(90,10,1000);        % start row index
 C = randi(totmods,numselect,1); % column index
 D = A(B:B+9,C);

The resulting matrix D should be the 10x1000 matrix wherein each column is a selection of 10x1 data starting from the index B for each column in A. However, the code does not work correctly. The code uses only B(:,1) as the row index for all the elements gathered from A in the line "D = A(B:B+9,C);"

How can this be corrected without using a loop?

[Guillaume]

The code uses only B(:,1) is slightly incorrect. The code actually only uses B(1, 1) because that's the documented behaviour of the colon operator:

   If you specify nonscalar arrays, MATLAB interprets j:i:k as j(1):i(1):k(1)

I have no idea which values of B (considering it is 10x1000 matrix) you're trying to use for each row of your D output.

Answer 1

[Thorsten]

You only need numelselect values of rows:

 B = randi(90,numselect,1);

Then use

 for i = 1:numel(B), D(:,i) = A(B(i):B(i)+9, C(i)); end

[Christopher]

Sorry, you are right about the variable B. I mistranscribed when converting my complex code to this simple one.

However, your suggested loop is precisely what I am using at the moment, but it is extraordinarily slow because numel(B) is actually very large (~10^7).

Is there a way to vectorize this operation?

[Thorsten]

Try

 D = A(bsxfun(@plus, bsxfun(@plus, B, 0:9), (C-1)*size(A,1))');

This computes the linear indices into A based on B and C.

[Christopher]

This has worked well, thanks

Answer 2

[Guillaume]

totmods = 10000;
numselect = 1000;
A = reshape(1:100*totmods, 100, totmods);      

B = randi(size(A, 1) - 9, numselect, 1); %column vector
C = randi(totmods, 1, numselect);  %row vector to make it easier later
rowmatrix = bsxfun(@plus, B, 0:9)'; %matrix of row indices
D = A(sub2ind(size(A), rowmatrix, repmat(C, 10, 1)))

Answer 3

[Stephen23]

Preallocate the output before allocating to it:

http://www.mathworks.com/help/matlab/matlab_prog/preallocating-arrays.html

http://blogs.mathworks.com/loren/2012/11/29/understanding-array-preallocation/

As numselect gets larger this will have a very significant effect on the calculation speed. Try something this:

B = randi(90,numselect,1);
D = nan(10,numel(B));
for k = 1:numel(B)
    D(:,k) = A(B(k):B(k)+9, C(k));
end