Documentaton - GENERICS - SVMKERNELTEST

Y = SVMKERNELTEST(CL, K) test the SVM CL on the testing data K. CL is a SVM obtained by SVMKERNELTRAIN(). K is the kernel matrix, training vectors along the columns and test vectors along the rows.

[Y, DEC] = SVMKERNELTEST(...) returns also the value DEC of the decision function(s) evaluated at each test vector. In order to parse this value, you must refer to the LAB = CL.LABELS vector. For a binary classifier LAB has two entries and DEC is a column vector, with one entry per test vector. Each value represent the confidence that the corresponding test vector class is LAB(1) as opposed to LAB(2).

For a multiclass classification problem with L classes, LAB has L entries DEC is a matrix. Each row contains the decision values of the L(L-1)/2 classification subproblems LAB(1) vs LAB(2), LAB(1) vs LAB(3) and so on. So for instance the second element of each row is the confidence that the correspdonging test vector class is LAB(12 as opposed to LAB(3).

DEC is the value of the decision functions with the bias removed (the bias is equal to - CL.LIBSVM_CL.RHO).

[Y, DEC, ACC] = SVMKERNELTEST(...) returns also the estimated prediction accuracy (see the Labels option below).

REMARK. When caluclating K, the order of the training vectors must match the order used to train the SVM (see SVMKERNELLEARN()).

Options:

Labels [[]]

  Specify test labels (used only to compute accuracy).

Probability [0]

  Return the probability instead of the decision function

Verbosity [0]

  Set verbosity level.