%WEAKC Weak Classifier
%
% [W,V] = WEAKC(A,ALF,ITER,CLASSF)
% [W,V] = A*WEAKC([],ALF,ITER,CLASSF)
% [W,V] = A*WEAKC(ALF,ITER,CLASSF)
%
% INPUT
% A Dataset
% ALF Fraction or number of objects to be used for training, see
% GENDAT. Default: one object per class. For ALF is integer, ALF
% objects per class are generated.
% ITER Number of trials
% CLASSF untrained classifier, default NMC
%
% OUTPUT
% W Best classifier over ITER runs
% V Cell array of all classifiers
% Use VC = stacked(V) for combining
% VC Combined set of classifiers
%
% DESCRIPTION
% WEAKC uses subsampled versions of A for training. Testing is done
% on the entire training set A. The best classifier is returned in W.
%
% VC = WEAKC(A,ALF,ITER,CLASSF,1)
%
% Combines all classifiers as a stacked combiner in VC.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% MAPPINGS, DATASETS, NMC, GENDAT, STACKED
% 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 [w,v] = weakc(varargin)
% INITIALISATION
argin = shiftargin(varargin,'scalar',1);
argin = setdefaults(argin,[],1,1,0,0);
if mapping_task(argin,'definition')
w = define_mapping(argin,'untrained','Weak');
% TRAINING
elseif mapping_task(argin,'training') % Train a mapping.
[a,n,iter,r,s] = deal(argin{:});
if isscalar(n) && n >= 1
n = n*ones(1,getsize(a,3));
end
v = {};
emin = 1;
for it = 1:iter % Loop
b = gendat(a,n); % subsample training set
if ~ismapping(r) % select classifier and train
if r == 0 % be consistent with old classfier selection
ww = nmc(b);
elseif r == 1
ww = fisherc(b);
elseif r == 2
ww = udc(b);
elseif r == 3
ww = qdc(b);
else
error('Illegal classifier requested')
end
else
if ~isuntrained(r)
error('Input classifier should be untrained')
end
ww = b*r;
end
v = {v{:} ww}; % store all classifiers in v
% select best classifier and store in w
e = a*ww*testc;
if e < emin
emin = e;
w = ww;
end
end
if s == 1
w = stacked(v);
end
end
return