unique.m

matlab6.5

function [b,ndx,pos] = unique(a,flag)
%UNIQUE Set unique.
%   UNIQUE(A) for the array A returns the same values as in A but
%   with no repetitions.  A will also be sorted.  A can be a cell
%   array of strings.
%
%   UNIQUE(A,'rows') for the matrix A returns the unique rows of A.
%
%   [B,I,J] = UNIQUE(...) also returns index vectors I and J such
%   that B = A(I) and A = B(J) (or B = A(I,:) and A = B(J,:)).
%   
%   See also UNION, INTERSECT, SETDIFF, SETXOR, ISMEMBER.

%   Copyright 1984-2002 The MathWorks, Inc. 
%   $Revision: 1.24 $  $Date: 2002/02/25 09:40:00 $

%   Cell array implementation in @cell/unique.m

nIn = nargin;

if nIn < 1
  error('Not enough input arguments.');
elseif nIn > 2
  error('Too many input arguments.');
end

if nIn == 1
  flag = [];
end

rows = size(a,1);
cols = size(a,2);

rowvec = (rows == 1) && (cols > 1); 

numelA = prod(size(a));
nOut = nargout;

if isempty(flag)
  
  % Handle empty: no elements.
  
  if (numelA == 0)
    % Predefine b to be of the correct type.
    b = a([]);
    if max(size(a)) > 0
      b = reshape(b,0,1);
      ndx = zeros(0,1);
      pos = zeros(0,1);
    else
      ndx = [];
      pos = [];
    end
    return
      
  elseif (numelA == 1)
    % Scalar A: return the existing value of A.
    b = a; ndx = 1; pos = 1;
    return
    
    % General handling.  
  else
    
    % Convert to columns
    a = a(:);
    
    % Convert to double array for purposes of faster sorting.
    % Additionally, UNIQUE calls DIFF, which requires double input.
    
    whichclass = class(a);    
    isdouble = strcmp(whichclass,'double');
    
    if ~isdouble
      a = double(a);
    end
    
    % Sort if unsorted.  Only check this for long lists.
    
    checksortcut = 1000;
    
    if numelA <= checksortcut || ~(issorted(a))
      if nOut <= 1
        b = sort(a);
      else
        [b,ndx] = sort(a);
      end
    else
      b = a;
      if nOut > 1
        ndx = (1:numelA)';  % If presorted, indices are 1,2,3,...
      end
    end
    
    % d indicates the location of non-matching entries.
    
    db = diff(b);
    
    % Since DIFF returns NaN in both Inf and NaN cases, 
    % use slower method of detection if NaN's detected in DIFF(b).
    % After sort, Infs or NaNs will be at ends of list only.
    
    if (isnan(db(1)) || isnan(db(numelA-1)))
      d = b(1:numelA-1) ~= b(2:numelA);
    else
      d = db ~= 0;
    end
    
    d(numelA,1) = 1;    % Final element is always member of unique list.
    
    b = b(d);         % Create unique list by indexing into sorted list.
    
    if nOut == 3
      pos = cumsum([1;full(d)]);  % Lists position, starting at 1.
      pos(numelA+1) = [];         % Remove extra element introduced by d.
      pos(ndx) = pos;             % Re-reference POS to indexing of SORT.
    end
    
    % Create indices if needed.
    if nOut > 1
      ndx = ndx(d);
    end
    
    % Re-convert to correct output data type using FEVAL.
    if ~isdouble
      b = feval(whichclass,b);
    end
  end
  
  % If row vector, return as row vector.
  if rowvec
    b = b.';
    if nOut > 1
      ndx = ndx.';
      if nOut > 2
        pos = pos.';
      end 
    end
  end
  
else    % 'rows' case
  if ~strcmpi(flag,'rows')
    error('Unknown flag.');
  end
  
  % Handle empty: no rows.
  
  if (rows == 0)
    % Predefine b to be of the correct type.
    b = a([]);
    ndx = [];
    pos = [];
    b = reshape(b,0,cols);
    if cols > 0
      ndx = reshape(ndx,0,1);
    end
    return
    
    % Handle scalar: one row.    
    
  elseif (rows == 1)
    b = a; ndx = 1; pos = 1;
    return
  end
  
  % General handling.
  % Conversion to double not done: SORTROWS is slower for doubles
  % than other types.
  
  if nOut > 1 
    [b,ndx] = sortrows(a);
  else
    b = sortrows(a);
  end
  
  % d indicates the location of non-matching entries.
  
  d = b(1:rows-1,:)~=b(2:rows,:);
  
  % d = 1 if differences between rows.  d = 0 if the rows are equal.
  
  d = any(d,2);
  d(rows,1) = 1;      % Final row is always member of unique list.
  
  b = b(d,:);         % Create unique list by indexing into sorted list.
  
  % Create position mapping vector using CUMSUM.
  
  if nOut == 3
        pos = cumsum([1;full(d)]);  % Lists position, starting at 1.
        pos(rows+1) = [];           % Remove extra element introduced by d.
        pos(ndx) = pos;             % Re-reference POS to indexing of SORT.
  end
  
  % Create indices if needed.
  if nOut > 1
    ndx = ndx(d);
  end
end