Introduction to Pattern Recognition: A Matlab Approach

48 CHAPTER 2 Classifiers Based on Cost Function Optimization

Table 2.4 Results for Various Values of C in Example 2.4.1

C = 0.1 C = 0.2 C = 0.5 C = 1 C = 2 C = 20

No.supportvectors8261443731 25

Training error 2.25% 2.00% 2.00% 2.25% 3.25% 2.50%

Test error 3.25% 3.00% 3.25% 3.25% 3.50% 3.50%

Margin 0.9410 0.8219 0.7085 0.6319 0.6047 0.3573

global figt4

figt4=2;

svcplot_book(X1',y1',kernel,kpar1,kpar2,alpha,-w0)

(c) To count the support vectors, type

sup_vec=sum(alpha>0)

(d) To compute the margin, type

marg=2/sqrt(sum(w.ˆ2))

The results of these experiments are shown in Table 2.4. It is readily seen that the margin of the

solution increases as C decreases.