GridGain Developers Hub

SVM Multi-class Classification

Multiclass SVM aims to assign labels to instances by using support vector machines, where the labels are drawn from a finite set of several elements.

The implemented approach for doing so is to reduce the single multiclass problem into multiple binary classification problems via one-versus-all.

The one-versus-all approach is the process of building binary classifiers which distinguish between one of the labels and the rest.

Model The model keeps the pairs <ClassLabel, SVMLinearBinaryClassificationModel> and it enables a prediction to be made for a given vector of features, in the following way:

SVMLinearMultiClassClassificationModel model = ...;

double prediction = model.predict(observation);


Presently, Ignite supports the following parameters for SVMLinearMultiClassClassificationTrainer:

  • amountOfIterations - amount of outer SDCA algorithm iterations. (default value: 200)

  • amountOfLocIterations - amount of local SDCA algorithm iterations. (default value: 100)

  • lambda - regularization parameter (default value: 0.4)

All properties will be propagated for each pair one-versus-all SVMLinearBinaryClassificationTrainer.

// Set up the trainer
SVMLinearMultiClassClassificationTrainer trainer = new SVMLinearMultiClassClassificationTrainer()

// Build the model
SVMLinearMultiClassClassificationModel mdl =