%FINDCLASSES Fixed mapping for finding object indices of all classes
%
% C = FINDCLASSES(A,NAME)
% C = A*FINDCLASSES([],NAME)
% C = A*FINDCLASSES(NAME)
%
% INPUT
% A Dataset
% NAME Integer: Index of desired labeling, see GETLABLISTNAMES
% String: Name of desired labeling, see GETLABLISTNAMES
% Default: actual LABLIST
%
% OUTPUT
% C Cell array, C{I} contains all object indices to class I.
%
% DESCRIPTION
% C is a cell array such that C{I} contains all object indices to class I.
% A(C{I},:) is a subset of A that contains class I only. Use REMCLASS to
% remove empty classes.
%
% The order of classes can be found by CLASSNAMES. If the name of the class
% is known then GETCLASSI can be used to retrieved its index.
%
% If for A multiple sets of class names (label lists, see MULTI_LABELING)
% are defined, the desired label list can be set by NAME. See also
% GETLABLISTNAMES.
%
% If A has N classes then C has N+1 cells. In the last cell the indices to
% all unlabeled samples are collected.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% DATASETS, SELCLASS, CLASSNAMES, GETCLASSI, REMCLASS, GETLABLISTNAMES,
% MULTI_LABELING
% Copyright: R.P.W. Duin, r.p.w.duin@37steps.com
function b = findclasses(varargin)
argin = shiftargin(varargin,{'char','scalar'});
argin = setdefaults(argin,[],[]);
if mapping_task(argin,'definition')
b = define_mapping(argin,'fixed');
b = setname(b,'Image stretch');
else
[a,name] = deal(argin{:});
if ~isempty(name)
a = changelablist(a,name);
b = feval(mfilename,a,clas);
else
nlab = getnlab(a);
[m,k,c] = getsize(a);
b = cell(1,c+1);
n = [0 classsizes(a)];
n = [m-sum(n) n(2:c+1)];
s = zeros(1,c+1);
for j=1:c+1
b{j} = zeros(1,n(j));
end
for j=1:m
p = nlab(j)+1;
s(p) = s(p)+1;
b{p}(s(p)) = j;
end
b_unlabeld = b(1);
b(1:c) = b(2:c+1);
if isempty(b_unlabeld{1})
b = b(1:c);
else
b(c+1) = b_unlabeld;
end
end
end