How to remove repeating rows without unique function?

I have a 624x2 matrix of ordered pairs, of which there are some repeating rows that I must remove. Is there a way to remove these repeating rows without using the unique function? Maybe using a for loop? The reason I can't use the unique function is because when I implement this matlab script in Simulink's Matlab Function block, the matrix must be sorted first before using the unique function. I can't sort the matrix because I will lose my ordered pairs and sortrows doesn't solve the issue either because I get the same error from Simulink.

Answer 1

[Cedric]

Use the 'stable' option:

 >> A = randi( 3, 10, 2 )
 A =
     2     3
     1     1
     3     1
     3     1
     3     1
     3     3
     3     3
     2     1
     2     3
     1     1

 >> unique( A, 'rows' )
 ans =
     1     1
     2     1
     2     3
     3     1
     3     3

 >> unique( A, 'rows', 'stable' )
 ans =
     2     3
     1     1
     3     1
     3     3
     2     1

[Andrew Poissant]

I have tried using the stable option and am still getting the errror "The second operand is not sorted in ascending order. Use SORT first" in Simulink. The second operand is referring to my unique function call

[Cedric]

I had never seen that. Here is another approach:

 >> A = randi( 3, 10, 2 )
     2     1
     2     3
     3     2
     3     1
     1     1
     2     2
     2     3
     2     2
     3     2
     3     1

 >> [A_s, id_s] = sortrows( A ) ;
    [~, id_u] = unique( A_s, 'rows' ) ;  % Passing sorted A to UNIQUE.
    A_u = A(sort( id_s(id_u) ), :)

which outputs

 A_u =
     2     1
     2     3
     3     2
     3     1
     1     1
     2     2

and just to check:

 >> isequal( A_u, unique( A, 'rows', 'stable' ))
 ans =
  logical
   1

So that seems to be working.

Note that Simulink users may know a better approach and/or why you seem to have to sort your data before you can use UNIQUE, because I have no clue!

[Andrew Poissant]

It is still giving me the same error. Sortrows does not work for sorting, Simulink needs the function SORT used for this. Is there any way to remove repeating rows without using unique at all?

[Cedric]

It would be better if a Simulink specialist would kick in and explain how this can be done properly in Simulink. I can only find "tricks" to emulate what you don't seem to have.

Here is another one:

 >> A = randi( 3, 10, 2 )
 A =
     3     1
     3     3
     1     3
     3     2
     2     3
     1     1
     1     2
     2     3
     3     3
     3     3

 >> checkFun = @(rowId) ~any(all(bsxfun(@eq, A(rowId,:), A(1:rowId-1,:)), 2)) ;
    select   = [true, arrayfun(checkFun, 2:size(A, 1))] ; 
    A_u      = A(select, :)

which outputs:

 A_u =
     3     1
     3     3
     1     3
     3     2
     2     3
     1     1
     1     2

and to check:

 >> isequal( A_u, unique( A, 'rows', 'stable' ))
 ans =
  logical
   1

[Andrew Poissant]

Yes, that would be nice if someone versed in simulink could offer advice, but I appreciate your help. Unfortunately, Simulink matlab function block can't take function handles or use arrayfun either. It is extremely annoying how limited the block is

[Cedric]

Then there is the good ol' loop:

 A = randi( 3, 10, 2 ) ;

 nRows = size(A, 1) ;
 keep  = true(nRows, 1) ;
 for rowId = 2 : nRows
    if any(all(repelem(A(rowId,:),rowId-1, 1) == A(1:rowId-1,:), 2))
        keep(rowId) = false ;
    end
 end
 A_u = A(keep,:) ;

Test:

 >> isequal(A_u, unique(A, 'rows', 'stable'))
 ans =
  logical
   1

If REPELEM is not supported, replace by REPMAT. If REPMAT not supported either, we can build the darn matrix using a product.

[Andrew Poissant]

It worked! Thank goodness. Really appreciate all of the help you gave me, you rock!

[Cedric]

My pleasure. The amount of unsupported features seems to be a real pain, good luck with the rest!

Answer 2

[Jan]

I still do not understand, why you cannot simply sort the data. But this might be useful:

!!! UNTESTED !!!

function Data = UniqueRows2(Data)
nRow = size(Data, 1);
keep = true(nRow, 1);
for k1 = 1:nRow
  if keep(k1)
    for k2 = k1 + 1:nData
      if keep(k2)
        % For rows of length 2:
        keep(k2) = (Data(k2, 1) ~= Data(k1, 1)) | (Data(k2, 2) ~= Data(k1, 2));
        % General row length:
        % keep(k2) = any(bsxfun(@ne, Data(k2, :), Data(k1, :)), 2);
      end
    end
  end
end
Data = Data(keep, :);
end

This might be faster with vectorization:

function Data = UniqueRows(Data)
nRow = size(Data, 1);
keep = true(nRow, 1);
for k1 = 1:nRow
  if keep(k1)
    keep(k1 + 1:nData) = and(keep(k1 + 1:nData), ...
        any(bsxfun(@ne, Data(k2, :), Data(k1, :)), 2));
  end
end
Data = Data(keep, :);
end

[Andrew Poissant]

Thanks! This also worked well for me!