%TESTD Find classification error of dataset applied to given classifier
%
% [E,C] = TESTD(A,W)
% [E,C] = TESTD(A*W)
% E = A*W*TESTD
%
% INPUT
% A Dataset
% W Classifier
%
% OUTPUT
% E Fraction of labeled objects in A that is erroneously classified
% C Vector with numbers of erroneously classified objects per class.
% They are sorted according to A.LABLIST
%
% DESCRIPTION
% This routine performs a straightforward testing of the labeled objects
% in A by the classifier W. Unlabeled objects are neglected. Just a
% fraction of the erroneously classified objects is returned. Not all
% classes have to be present in A and class priors are neglected. TESTD
% thereby essentially differs from TESTC which tests the classifier instead
% of the dataset. Labels found for the individual objects by the classifier
% W may be retrieved by LABELD
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% DATASETS, MAPPINGS, TESTD, LABELD
% 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 [error,class_errors] = testd(a,w)
if nargin < 2, w = []; end
if nargin < 1 | isempty(a)
error = prmapping(mfilename,'fixed',{w});
error = setname(error,'testd');
class_errors = [];
elseif isempty(w)
s = classsizes(a);
c = getsize(a,3);
class_errors = zeros(1,c);
n = 0;
for j=1:c
b = seldat(a,j);
if ~isempty(b)
class_errors(j) = nlabcmp(getlabels(b),b*labeld);
end
n = n+size(b,1);
end
error = sum(class_errors)/n;
else
[error,class_errors] = testd(a*w);
end
return