Model Training

Building Models from Command Line

All learning algorithms in MLTK inherits mltk.predictor.Learner class. To build models from command line, use the following command (using LassoLearner as example):

$ java mltk.predictor.glm.LassoLearner

It should output a message like this:

Usage: mltk.predictor.glm.LassoLearner
-t	train set path
[-g]	task between classification (c) and regression (r) (default: r)
[-r]	attribute file path
[-o]	output model path
[-V]	verbose (default: true)
[-m]	maximum num of iterations (default: 0)
[-l]	lambda (default: 0)

Building Models in Java Code

Using LassoLearner as example, the following code builds a L1-regularized classification model:

LassoLearner learner = new LassoLearner();
learner.setTask(Task.CLASSIFICATION);
learner.setLambda(0.01);
learner.setMaxNumIters(100);
		
GLM glm = learner.build(trainSet);

It is recommended to use build method to train the model since it is the most flexible method and other meta learning algorithms such as BaggedEnsembleLearner will call this method to build models. Some learning algorithms will have other public methods for training to avoid pre-setting all parameters. For example, the following code does the same thing:

LassoLearner learner = new LassoLearner();
GLM glm = learner.buildClassifier(trainSet, 100, 0.01);