STATSSVC

Stats Support Vector Classifier (Matlab Stats Toolbox)

    W = STATSVC(A,KERNEL,C,OPTTYPE)
    W = A*STATSVC(KERNEL,C,OPTTYPE)
    D = B*W

Input
A	A PRTools dataset used fro training
KERNEL	Untrained mapping to compute kernel by A*(A*KERNEL) during  training, or B*(A*KERNEL) during testing with dataset B.  Default: linear kernel (PROXM('p',1))
C	Regularization ('boxconstraint' in SVMTRAIN)
OPTTYPE	Desired optimizer, 'SMO' (default), 'QP' or 'LS'.
B	PRTools dataset used for testing

Output
W	Mapping: Support Vector Classifier
D	PRTools dataset with classification results

Description

This is the PRTools interface to the support vector classifier SVMTRAIN in Matlab's Stats toolbox. Like PRTools SVC it is a two-class  discriminant that also can be used for multi-class problems by an  internal call to MCLASSC. SVMTRAIN uses the Sequential Minimal  Optimization (SMO) method and thereby implements the L1 soft-margin SVM classifier. See SVMTRAIN for more details.

Non-linear kernels have to be supplied by kernel procedures like PROXM.  It is assumed that V = A*KERNEL generates the trained kernel and B*V the  kernel matrix with size(B,1) rows and size(A,1) columns. Forthe radial  basis kernel use PKSTATSSVC and RBSTATSSVC

Like for all other PRTools classifiers, a new dataset B can be classified  by D = B*W. The classifier performance can be measured by D*TESTC and the  resulting labels by D*LABELD. D is a dataset with for every class a  column. The values can be considered as class confidences or classifier  condaitional posteriors.

STATSSVM is basically a two-class classifier. Multiclass problems are  internally solved using MCLASSC resulting in a base classifier per class.  The final result may be improved significantly by using a non-linear  trained combiner, e.g. by calling W = A*(STATSSVM*QDC([],[],1e-6);

Alternative SVM classifiers in PRTools are based on SVC and LIBSVC.

See also

datasets, mappings, svmtrain, svc, libsvc, mclassc, pkstatssvc, rbstatssvc, qdc, testc, labeld,