%TESTK Error estimation of the K-NN rule
%
% E = TESTK(A,K,T)
%
% INPUT
% A Training dataset
% K Number of nearest neighbors (default 1)
% T Test dataset (default [], i.e. find leave-one-out estimate on A)
%
% OUTPUT
% E Estimated error of the K-NN rule
%
% DESCRIPTION
% Tests a dataset T on the training dataset A using the K-NN rule and
% returns the classification error E. In case no set T is provided, the
% leave-one-out error estimate on A is returned.
%
% The advantage of using TESTK over TESTC is that it enables leave-one-out
% error estimation. However, TESTK is based on just counting errors and
% does not weight with testobject priors.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% DATASETS, KNNC, KNN_MAP, TESTC
% 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
% $Id: testk.m,v 1.4 2009/07/22 19:29:41 duin Exp $
function [e,labels] = testk(a,knn,t)
if (nargin < 2)
prwarning(2,'number of neighbours K not specified, assuming 1');
knn = 1;
end
% Calculate the KNN classifier.
a = seldat(a); % get labeled objects only
w = knnc(a,knn);
[m,k] = size(a);
nlab = getnlab(a);
lablist = getlablist(a);
if (nargin <= 2) % Leave-one-out error estimate.
d = knn_map([],w); % Posterior probabilities of KNNC(A,KNN).
[dmax,J] = max(d,[],2); % Find the maximum.
e = nlabcmp(J,nlab)/m; % Calculate error: compare numerical labels.
else % Error estimation on tuning set T.
[n,kt] = size(t);
if (k ~= kt)
error('Number of features of A and T do not match.');
end
d = knn_map(t,w); % Posterior probabilities of T*KNNC(A,KNN).
[dmax,J] = max(d,[],2); % Find the maximum.
% Calculate error: compare full labels.
e = nlabcmp(getlab(t),lablist(J,:))/n;
end
labels = lablist(J,:);
return