How can I index a multidimensional array with another multidimensional array containing indices into one of the dimensions of the first array?

Hello everyone.

I have the following problem. Let me give an example:

Suppose I have a 4-D array A of real numbers of 10000x10x10x2. I also have a 4-D array of indices into the first dimension of A, call it I, which is of the size 4500x10x10x2. (so all values in I are integers between 1 and 10000). I need an output array B of the size 4500x10x10x2 in which the first dimension would be the indexed dimension of A applied element by element for the other 3 dimensions. What is the efficient way of doing it? A(I) does not work.

There is a also constraint on I is that every vector along the 1-st dimension (4500) has consecutive integer numbers, so, say I(:,1,1,1) could be, for example, 1,2,3,...4500; or 1000,1001,....5499. (thus I is uniquely defined by 200 values)

This can be solved with a triple nested loop over dimensions 2,3, and 4, but it is slow and I need to repeat this process many times. I can also reshape both matrices to 10000x200 and 4500x200, respectively, and use a single "for" loop over the 2-nd dimension, but still have to loop over. Is there an elegant solution for this? Maybe sub2ind can be used, but I don't see how.

Thanks a lot.

Answer 1

[Roger Stafford]

 [n1,n2,n3,n4] = size(I);
 I2 = repmat(size(A,1)*(0:n2*n3*n4-1),n1,1);
 B = reshape(A(I(:)+I2(:)),size(I));

[Ilya Udovydchenkov]

Thank you so much. Works great!