PKSVC

Automatic radial basis SVM, using NUSVC and the Parzen kernel

    [W,KERNEL,NU,C] = PKSVC(A,ALF)
    [W,KERNEL,NU,C] = A*PKSVC([],ALF)
    [W,KERNEL,NU,C] = A*PKSVC(ALF)

Input
A	Dataset
ALF	Parameter, default 1

Output
W	Mapping: Radial Basis Support Vector Classifier
KERNEL	Untrained mapping, representing the optimised kernel
NU	Resulting value for NU from NUSVC (W = NUSVC(A,KERNEL,NU))
C	Resulting value for C (W = SVC(A,KERNEL,C)

Description

This routine provides a radial basis support vector classifier based on  NUSVC (which estimates NU using the leave-one-out 1NN error) and  estimates the kernel width SIGMA by the the value found by PARZENC. The  kernel width used is ALF*3*SQRT(2)*SIGMA. This is much faster than the  gridsearch used by RBSVC and performs about equally well.

See also

datasets, mappings, svc, nusvc, rbsvc, parzenc, libsvc,