Scoring Algorithms¶
CPSign has two different machine learning algorithms implemented, LibLinear and LibSvm, and these are used as scoring algorithms for the Predictor used.
Table of Contents
MLAlgorithm wrapping¶
CPSign wraps LibLinear and LibSvm java implementation in the classes LibLinear and LibSvm
to provide a generic interface to the user. Instantiation can be done either directly by the constructors
of the classes or by the CPSignFactory: using:
// Direct instantiation
LibLinear libLin = new LibLinear(new Parameter(SolverType.L1R_L2LOSS_SVC, c, eps));
// Using default parameters in CPSign
CPSignFactory factory = ...; // Instantiate
LibLinear libLin = factory.createLibLinearClassification(); // classification
LibLinear libLin = factory.createLibLinearRegression(); // regression
// Direct instantiation of LibSvm
LibSvm svm = new LibSvm(new svm_parameter()); // set your parameters on the svm_parameter object
// Using factory
factory.createLibSvmClassification();
factory.createLibSvmRegression();
Setting custom parameters¶
By default CPSign uses parameters for LibLinear and LibSVM that has been found to produce good results together with signatures descriptors [6]. However, it is possible to to set all LibLinear and LibSVM parameters programmatically through the API. This is mostly important when performing predictions where data come from other sources than derived from signatures. Setting parameters are as straightforward as:
// For LibLinear
// Create a LibLinear object either with classification or regression settings
LibLinear liblin = factory.createLibLinearClassification();
LibLinearParameters params = liblin.getParameters();
// Either set parameters one by one
params.setC(newC);
params.setEpsilon(newEpsilon);
// Or create new parameters from scratch
Parameter liblinParams = new Parameter(SolverType.L1R_LR, 100, 0.5);
liblin.setParameters(liblinParams);
// For LibSvm
// Create a LibSvm object either for classification or regression
LibSvm impl = factory.createLibSvmClassification();
LibSvmParameters params = impl.getParameters();
// Or create new parameters from scratch
svm_parameter svmParams = new svm_parameter();
svmParams.C = 100;
svmParams.eps = 0.5;
liblin.setParameters(svmParams);