Grouping matrix rows in terms of one column

Question

First of all it is really hard for me to describe the problem really good but I'll try.

Say that we have matrix A

A = [23 1;
     45 1
     78 1
     86 1
     98 2
     1  2
     23 2
     14 3
     15 4
     85 4]

What I want as an output is

    B{1} = [23,45,78,86]
    B{2} = [98,1,23]
    B{3} = [14]
    B{4} = [15,85]

Bear in mind that the original A is a huge matrix, and I do not wanna do this with for loops. I would like to use functions that uses parallel processing.

Answer 1

You could use accumarray here:

B = accumarray(A(:,2),A(:,1),[],@(x){x},{});

If you know that A is sorted, and that there is no missing entry from the second column, you can also use mat2cell:

counts = histc(A(:,2),unique(A(:,2)));
B = mat2cell(A(:,1),counts);

Answer 2

Here's a simple way to do it. It uses a loop, but should be pretty quick because the cell array B is pre-allocated using the reverse index trick.

for key = fliplr(unique(A(:,2)'))
    ndx = A(:,2) == key;
    B{key} = A(ndx,1)';
end