edu.umass.cs.mallet.base.classify
Class WinnowTrainer

java.lang.Object
  edu.umass.cs.mallet.base.classify.ClassifierTrainer
      edu.umass.cs.mallet.base.classify.WinnowTrainer

public class WinnowTrainer

extends ClassifierTrainer

An implementation of the training methods of a Winnow2 on-line classifier. Given an instance xi, the algorithm computes Sum(xi*wi), where wi is the weight for that feature in the given class. If the Sum is greater than some threshold theta, then the classifier guess true for that class. Only when the classifier makes a mistake are the weights updated in one of two steps: Promote: guessed 0 and answer was 1. Multiply all weights of present features by alpha. Demote: guessed 1 and answer was 0. Divide all weights of present features by beta. Limitations: Winnow2 only considers binary feature vectors (i.e. whether or not the feature is present, not its value).

Constructor Summary

WinnowTrainer()
Default constructor.

WinnowTrainer(double a, double b)
Sets alpha and beta and default value for theta

WinnowTrainer(double a, double b, double nfact)
Sets alpha, beta, and nfactor

Method Summary

Classifier

train(InstanceList trainingList, InstanceList validationList, InstanceList testSet, ClassifierEvaluating evaluator, Classifier initialClassifier) Trains winnow on the instance list, updating weights according to errors

Methods inherited from class edu.umass.cs.mallet.base.classify.ClassifierTrainer

main, toString, train, train, train, train

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

WinnowTrainer

public WinnowTrainer()

Default constructor. Sets all features to defaults.

WinnowTrainer

public WinnowTrainer(double a,
                     double b)

Sets alpha and beta and default value for theta

Parameters:

a - alpha value

b - beta value

WinnowTrainer

public WinnowTrainer(double a,
                     double b,
                     double nfact)

Sets alpha, beta, and nfactor

Parameters:

a - alpha value

b - beta value

nfact - nfactor value

Method Detail

train

public Classifier train(InstanceList trainingList,
                        InstanceList validationList,
                        InstanceList testSet,
                        ClassifierEvaluating evaluator,
                        Classifier initialClassifier)

Trains winnow on the instance list, updating weights according to errors

Specified by:

train in class ClassifierTrainer

Parameters:

trainingList - examples used to set parameters.

validationList - examples used to tune meta-parameters. May be null.

testSet - examples not examined at all for training, but passed on to diagnostic routines. May be null.

initialClassifier - training process may start from here. The parameters of the initialClassifier are not modified. May be null.

Returns:

Classifier object containing learned weights