find reverses the dimension of rows and column output vectors when input is a vector

Hi together,

I observed some behavior of the find function, where I am not sure if this is intended or simply a bug.

When calling find on an 2-Dimensional array (with both dimensions are greater than 1) the returned row and column indices by find are a Nx1 vectors. You can concatenate them to get a Nx2 matrix with the indices like in the following example:

arr = [1 0 0; 0 1 1];
[row, col] = find(arr); % both are a Nx1 vector
indices = [row, col]; % this breaks if arr is 1x2

If now my input array has only one row (i.e. it is a horizontal vector), find still returns rows and column indices, but this time those are 1xN vectors and concatenating them in the same way like above with

indices = [row, col]

will not end up with the intended result. In my case those arrays are tables, where there are cases where the table has multiple lines and also cases where it only has one. The indices are needed for formatting specific cells in this table and it takes the format Nx2.

Do I now really need to add a check in my code wether the input array has only one line and then transpose the row and column vectors returned by find?

if size(arr, 1) == 1
    rows = rows';
    cols = cols';
end

Thanks for your help

Answer 1

[dpb]

It has always behaved to return the shape of the input -- a row vector input returns a row vector output. That is expected behavior throughout MATLAB and it would break a lot of code to change it, so that "ain't agonna happen!".

So, "Yes, Virginia", you do need to test for orientation to catenate the outputs if want them as column vectors -- or write to force them to be without the logical test as

>> [row, col] = find(arr(:).')
row =
     1     1     1
col =
     1     4     6
>> indices=[row(:) col(:)]
indices =
     1     1
     1     4
     1     6
>> 

This forces them to be column vectors when catenate and while not necessary in case they already are, doesn't hurt anything and saves the logical test.

This becomes a natural MATLAB idiom with experience...