Trainable automatic radial basis Support Vector Classifier
[W,KERNEL,NU,C] = RBSVC(A)
This routine computes a classifier by NUSVC using a radial basis kernel with an optimised standard deviation by REGOPTC. The resulting classifier W is identical to NUSVC(A,KERNEL,NU). As the kernel optimisation is based on internal cross-validation the dataset A should be sufficiently large. Moreover it is very time-consuming as the kernel optimisation needs about 100 calls to SVC.
If any class in A has less than 20 objects, the kernel is not optimised by a grid search but by PKSVM, using the Parzen kernel.
Note that SVC is basically a two-class classifier. The kernel may thereby be different for all base classifiers and is separately optimised for each of them.