Unique Rows for a cell array

Find unique rows of a cell array containing columns with strings or scalars, or N-D matrices

Description

unique(cA,'rows') does not work for cell arrays, the rows entry is ignored (at least as of 2009b). This function handles this case. It works for strings or scalars or both, but each column must contain only one or the other.

• MATLAB release: MATLAB 7.9 (2009b)

[Matt Fig]

This is much faster, and has no requirement as to types or ordering:

ii = 1;
while ii<size(A,1)
    tf = true(size(A,1),1);
    for jj = ii:size(A,1)
        if isequal(A(ii,:),A(jj,:)) && ii~=jj
            tf(jj) = 0;
        end
    end
    A = A(tf,:);
    ii = ii + 1;
end

For example, try it with:
A = {1,'red',magic(3);'blue',magic(4),'green';1,'red',magic(3)}

[Jim Hokanson]

Thanks for the comment Matt. The approach you suggest is definitely faster and more flexible for a small example like the one shown. Unfortunately the approach shown becomes nearly intractable for a larger number of rows when the number of unique rows is also large.

I tried a modified A matrix, that followed the rules outlined in the code. I updated the code tonight to begin to handle N-D matrices (currently they need to be of the same size, although I have a decent idea on how I can change the code to handle mismatches).

A = {1,'red', [3 2;4 5]; ...
    2,'green', [4 4;5 6]};

To get more samples I added the following code:
A = repmat(A,[500 1]); %I had 2 rows to begin with
A(:,1) = num2cell(rand(size(A,1),1));

The second line gives the variability and can be commented out to see how large an effect multiple output rows can have.

On my computer the calculation took 30ms for 1000 rows compared to 10 ms with your code. However, if I make the numeric entries random, my code takes about the same amount of time as it had before, but your code takes 2 seconds. For 10000 rows, this changes to approximately 200 seconds (I only it once and got 213) vs 0.2 seconds for my code.

Interestingly, one of the slowest aspects of this code turned out to be cell2mat, which is called by sortrows. I rewrote my own version based a newsthread online. It is a bit frustrating though that TMW doesn't have a really quick way of going from something like {1 2 3} to [1 2 3].

[epoxy patch]

thanks - exactly what i needed, and it works well

[Sriram]

is way faster than the code by Matt Fig for cell arrays of strings

[Nick]

Excellent, and works very fast even on 100,000+ rows.

[ALiveris]

Worked for me fast (1s) on a 50000x3 array of strings

[Nate]

Great job. Exactly what I was looking for.

[Ulrik]

 

[Ulrik]

input check not implemented well. eg. if nargin == 2 the it will fail with an error.

[Jessica Lam]