ECG-Kit 1.0

%BAGCC Combining classifier for classifying bags of objects 
%
%		DBAG = BAGCC(DOBJ,COMBC)
%		DBAG = DOBJ*BAGCC([],COMBC)
%		DBAG = DOBJ*BAGCC(COMBC)
%
% INPUT
%   DOBJ   Dataset, classification matrix, output of some base classifier
%   COMBC  Combiner, e.g. MAXC (default VOTEC)
%
% OUTPUT
%   DBAG   Dataset, classification matrix for the bags in DOBJ
%
% DESCRIPTION
% This routine combines object classification results of bags of objects 
% stored in DOBJ. It is assumed that the current labels of DOBJ are bag 
% identifiers and defining objects belonging to the same bag. Objects of 
% the same bag are combined by COMBC into a single classification result 
% and returned by DBAG. 
%
% DBAG gets as many objects as there are bags defined for DOBJ. Effectively
% the first object of every bag in DOBJ is replaced by the combined result
% and other objects of that bag are deleted. DBAG has the same feature 
% labels (most likely the class names) as DOBJ and stores as object 
% identifiers the bag identifiers stored in the label list of DOBJ. 
% A possible multi-labeling definition of DOBJ is preserved.
%
% This routine is called by BAGC where needed.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% MAPPINGS, DATASETS, BAGC, MULTI-LABELING

% Copyright: R.P.W. Duin, r.p.w.duin@37steps.com
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands

function [dbag,id] = bagcc(dobj,combc)

	argin = shiftargin(varargin,'prmapping');
  argin = setdefaults(argin,[],votec);
  if mapping_task(argin,'definition')
    out1 = define_mapping(argin,'untrained');
    out1 = setname(out1,'Bag combiner');
  else
    [dobj,combc] = deal(argin{:});

    % we should have a proper dataset
    isdataset(dobj);

    % the class names are the bag indentifiers
    bagnames = classnames(dobj);

    % retrieve datasize, and number of sets c
    [m,k,c] = getsize(dobj);

    % get number of objects for every set
    s = classsizes(dobj);

    % dobj is a classification matrix, so its features point to classes
    featlab = getfeatlab(dobj);

    % reserve spave for the result
    dbag = prdataset(zeros(c,k));

    % space the object identifiers of the first object per bag
    id = zeros(c,1);

    t = sprintf('Combining %i bags: ',c);
    prwaitbar(c,t);

    % run over all bags
    for j=1:c
      prwaitbar(c,j,[t int2str(j)]);

      % get the objects in the bag
      y = seldat(dobj,j);

      % the identifier of the first object
      id(j) = getident(y(1,:));

      %create a dataset with all objects in the bag concatenated horizontally
      y = +y';
      y = prdataset(y(:)');

      % give the the proper class labels
      y = setfeatlab(y,repmat(featlab,s(j),1));

      % now classifier combiners can be used
      dbag(j,:) = y*combc;

    end
    prwaitbar(0);

    % find the first objects of every set
    J = findident(dobj,id); 

    % and replace them by the bag combining result
    % so object labels become bag labels
    dbag = setdata(dobj(J,:),dbag);

    % give columns the classnames
    dbag = setfeatlab(dbag,featlab);

    % use set bag names as bag identifiers
    dbag = setident(dbag,bagnames);
  end