Row wise indexing of 3d array

I have a 3d array, for example:

(:,:,1)
 -0.1468   -0.4846   -0.3310
  0.3212   -0.4570    0.1491

(:,:,2)
 -0.3110   -0.3165    0.1256
  0.1868   -0.1315    0.2802

(:,:,3)
 -0.4189    0.2757   -0.0641
  0.4294   -0.0132   -0.0532

What I want to do is find all values of the last column (:,3,:) that is less than 0, and then set the corresponding rows to NaN. In the above case, the results should be:

(:,:,1)
 NaN        NaN       NaN
  0.3212   -0.4570    0.1491

(:,:,2)
 -0.3110    0.3165    0.1256
  0.1868   -0.1315    0.2802

(:,:,3)
  NaN       NaN       NaN
  NaN       NaN       NaN

I can achieve it in the 2d case using the following code:

array = rand(5,3)-0.5
array(find(array(:,3)<0),:) = NaN

But I struggle with the 3d case.

Answer 1

[Guillaume]

Note that in the 2D case you didn't need find (which only slows things done):

array(array(:, 3) < 0, :) = NaN;

For the ND case it's a bit more complicated since the result of the comparison is a matrix not a single dimension vector. One possibility:

array(repmat(array(:, 3, :) < 0, 1, size(array, 2), 1)) = NaN

[James Tursa]

On earlier versions of MATLAB that trailing argument of repmat will not work. So just this:

array(repmat(array(:, 3, :) < 0, 1, size(array, 2))) = NaN

[Oscar Frick]

This works perfectly as well as being effective. Some short testing I did seems to indicate that the version without the trailing argument has an extremely slight time advantage when having "small" arrays (on the format (n,3,4), small being size seems to be about n<10000000). For larger arrays the version with the trailing one starts getting ahead, getting a more significant efficiency advantage as the arrays grow bigger.

Answer 2

[Mark Saad]

You could try:

    [I,J,K] = ind2sub(size(array), find(array(:,:,:) < 0));
    ind = find(J == 3);
    I = I(ind);
    J = J(ind);
    K = K(ind);
    array(I,:,K) = NaN;

The first line gets the indices of every value that is less than 0. Lines 2-4 filter out any indices not in the third column. The last line sets the desired values to NaN.