edu.cmu.minorthird.classify.algorithms.linear
Class VitorBalancedWinnow

java.lang.Object
  edu.cmu.minorthird.classify.OnlineClassifierLearner
      edu.cmu.minorthird.classify.OnlineBinaryClassifierLearner
          edu.cmu.minorthird.classify.algorithms.linear.VitorBalancedWinnow

All Implemented Interfaces:

BinaryClassifierLearner, ClassifierLearner, java.io.Serializable, java.lang.Cloneable

public class VitorBalancedWinnow
extends OnlineBinaryClassifierLearner
implements java.io.Serializable

Created on Sep 22, 2005

Author:

Vitor R. Carvalho Implements the modified version of the Balanced Winnow algorithm (MBW), following the ideas described in "Improving Winnow for NLP Tasks: Voting Schemes, Regret Minimization and Online Feature Selection", Vitor Carvalho, William Cohen and Avrim Blum, CMU Technical Report 2006 A Modified version of the Balanced Winnow algorithm, as described in "Learning Quickly when Irrelevant Attributes Abound: a new linear-threshold algorithm", N. Littlestone, Machine Learning, 1988. One of the modifications is based on the "thick-Margin trick used in "Mistake-Driven Learning in Text Categorization", I. Dagan, Y. Karov, D. Roth, EMNLP, 1997 Additionally, it implements the Voting scheme (averaged hypothesis). Please check the "voted" parameter for that.

See Also:

Serialized Form

Nested Class Summary

class

VitorBalancedWinnow.MyClassifier

Constructor Summary

VitorBalancedWinnow()
Constructor

VitorBalancedWinnow(boolean voted)
Constructor: voted="true" (more stable) or "false"(more aggressive)

VitorBalancedWinnow(double alpha, double beta, boolean voted)

Method Summary

void	addExample(Example example2) Subclasses should use this method to perform whatever incremental update is needed after in response to a new example.
Classifier	getClassifier() Subclasses should use this method to return the current classifier.
void	reset() 'forget' everything about the last learning task, and start a new task.
java.lang.String	toString()
void	updateVotedHyperplane(int count)

Methods inherited from class edu.cmu.minorthird.classify.OnlineBinaryClassifierLearner

getBinaryClassifier, getSchema, setSchema

Methods inherited from class edu.cmu.minorthird.classify.OnlineClassifierLearner

completeTraining, copy, hasNextQuery, nextQuery, setInstancePool

Methods inherited from class java.lang.Object

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

Methods inherited from interface edu.cmu.minorthird.classify.ClassifierLearner

completeTraining, copy, hasNextQuery, nextQuery, setInstancePool

Constructor Detail

VitorBalancedWinnow

public VitorBalancedWinnow()

Constructor

VitorBalancedWinnow

public VitorBalancedWinnow(boolean voted)

Constructor: voted="true" (more stable) or "false"(more aggressive)

Parameters:

voted -

VitorBalancedWinnow

public VitorBalancedWinnow(double alpha,
                           double beta,
                           boolean voted)

Method Detail

reset

public void reset()

Description copied from class: OnlineClassifierLearner

'forget' everything about the last learning task, and start a new task. Subclasses need to implement this method

Specified by:

reset in interface ClassifierLearner

Specified by:

reset in class OnlineClassifierLearner

addExample

public void addExample(Example example2)

Description copied from class: OnlineClassifierLearner

Subclasses should use this method to perform whatever incremental update is needed after in response to a new example.

Specified by:

addExample in interface ClassifierLearner

Specified by:

addExample in class OnlineClassifierLearner

updateVotedHyperplane

public void updateVotedHyperplane(int count)

getClassifier

public Classifier getClassifier()

Description copied from class: OnlineClassifierLearner

Subclasses should use this method to return the current classifier.

Specified by:

getClassifier in interface ClassifierLearner

Specified by:

getClassifier in class OnlineClassifierLearner

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object