dd_ex1.m

SVDD的工具箱

% DD_EX1
%
% Example of the creation of a One-Class problem, and the solutions
% obtained by the Nearest Neighbor Data Description and the Support
% Vector Data Description. Furthermore, the ROC curve is plotted and
% the AUC-error and F1-performance is computed.

% Copyright: D.M.J. Tax, [e-mail]
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands

% generate normal classification problem: train data:
a = gendatb([30 30]);
% make the second class the target class and change the labels:
a = oc_set(a,'1');
% only use target class:
a = target_class(a);
% generate test data:
b = oc_set(gendatb(200),'1');

% first show a 2D plot:
figure(1); clf; hold on; H = [0;0];
h = scatterd(a);
V = axis; axis(1.5*V);

% train the individual data descriptions and plot them
% the error on the target class:
fracrej = 0.2;
% train the nndd:
w1 = nndd(a,fracrej);
% and plot the decision boundary:
h = plotc(w1,'k-');
H(1) = h(1);
% second, train the svdd:
w2 = svdd(a,fracrej,5);
% and also plot this:
h = plotc(w2,'r--');
H(2) = h(1);
legend(H,'NNdd','SVDD');
axis equal;
axis image;

% second show the ROC curves:
figure(2); clf;hold on; H=[0;0];
% the ROC for the Nearest Neighbor method:
e1 = dd_roc(b,w1);
% plot the ROC curve:
h = plotroc(e1,'k-'); H(1) = h(1);
% the area under the ROC curve:
auc_nn = dd_auc(e1)
% and the f_1 performance:
f1_nn = b*w1*dd_f1
% the ROC for the SVDD:
e2 = dd_roc(b,w2);
% also plot this:
h = plotroc(e2,'r--'); H(2) = h(1);
% and compute the area under the ROC:
auc_svdd = dd_auc(e2)
% and the f1 measure:
f1_svdd = b*w2*dd_f1;

legend(H,'NNdd','SVDD');