Find columns with same values in matrix

Hi,

I actually struggle on a fast and good solution on finding columns in an 2D matrix containing same values and returning the row index of those columns.

% Example:
A = [2,2,7,3;  ...
     8,7,8,10; ...
     4,3,5,2;  ...
     6,7,1,9];

Result should be 2 and 4 because in column 2 the number 7 occurs two times. Is there anything similar to unique or anything else? I try to avoid looping every column with unique since there will be a lot of matrices and bigger ones.

Thanks a lot!

[madhan ravi]

4 ???

[Manuel Schmidberger]

Yea because A(2,2) and A(4,2) are the same values in column2 and I need to know the rows where the same values occur.

At the end I got a lot of matrices with thunderstorm tracks which undergo merging and splittings and I need to find the mergings. Each track is stored in a row. Merged and splitted tracks have at some timesteps (= columns) the exact same coordinates. And the first timestep can be assumed to be the merging time.

Answer 1

[Adam Danz]

I changed your example matrix to include a variety of duplicates.

% Example:
A = [2,2,7,3,5,0;8,7,8,10,5,0;4,3,5,3,9,0;6,7,1,9,9,1];

A =
     2     2     7     3     5     0
     8     7     8    10     5     0
     4     3     5     3     9     0
     6     7     1     9     9     1

As you can see, columns 2, 4, 5 & 6 have duplicates and column 5 has 2 sets while column 6 has three of the same value.

% identify the duplicate values
dupIdx = splitapply(@(x){histc(x,unique(x))>1}, A, 1:size(A,2)); 
unqVals = splitapply(@(x){unique(x)}, A, 1:size(A,2)); 
dupVals = cellfun(@(x,y)x(y), unqVals, dupIdx, 'UniformOutput', false); 

% List the rows that contain duplicate values for each column
dupRows = splitapply(@(x,y) {find(ismember(x,[y{:}]))}, A, dupVals, 1:size(A,2)); 
% List the column numbers that contain duplicates
hasDup = find(~cellfun(@isempty, dupRows)); 

So, dupRows{2} will list the row numbers of column 2 that contain a duplicate. It's empty if col 2 has no duplicates.

hasDup is a vector of column numbers that contain a duplicate. It's empty if there are no duplicates in any column.

[UPDATE] For matlab releases before 2015b, here's the same method using cellfun() instead of splitapply()

% Example:
A = [2,2,7,3,5,0;8,7,8,10,5,0;4,3,5,3,9,0;6,7,1,9,9,1];

Acell = mat2cell(A,size(A,1),ones(1,size(A,2)));
dupIdx = cellfun(@(x){histc(x,unique(x))>1}, Acell);
unqVals = cellfun(@(x){unique(x)}, Acell); 
dupVals = cellfun(@(x,y)x(y), unqVals, dupIdx, 'UniformOutput', false); 
% List the rows that contain duplicate values for each column
dupRows = cellfun(@(x,y) find(ismember(x,y(:))), Acell, dupVals,'UniformOutput', false); 
% List the column numbers that contain duplicates
hasDup = find(~cellfun(@isempty, dupRows)); 

% Alternative
% hasDup = any(([ones(1,size(A,2));diff(sort(A))]~=0)==0);

[Manuel Schmidberger]

Hi

sorry was out of office a few days. Unfortunatly splitapply is not included in 2015a release or is it in a certain toolbox?

Thank you!

Cheers

[Adam Danz]

splitapply() came out in r2015b. I updated my solution to include a second method using cellfun() instead of splitapply(). It's the same logic; just adapted to cellfun().

[Manuel Schmidberger]

Yeah thanks a lot, that works now!

Really need to update the matlab release on cluster due to some really good new functions in newer releases!

Thanks!