SVM::crossvalidate

(PECL svm >= 0.1.0)

Test training params on subsets of the training data.

public float svm::crossvalidate ( array $problem, int $number_of_folds )

Crossvalidate can be used to test the effectiveness of the current parameter set on a subset of the training data. Given a problem set and a n "folds", it separates the problem set into n subsets, and the repeatedly trains on one subset and tests on another. While the accuracy will generally be lower than a SVM trained on the enter data set, the accuracy score returned should be relatively useful, so it can be used to test different training parameters.

Parameters:

problem

The problem data. This can either be in the form of an array, the URL of an SVMLight formatted file, or a stream to an opened SVMLight formatted datasource.

number_of_folds

The number of sets the data should be divided into and cross tested. A higher number means smaller training sets and less reliability. 5 is a good number to start with.

Returns:

The correct percentage, expressed as a floating point number from 0-1. In the case of NU_SVC or EPSILON_SVR kernels the mean squared error will returned instead.