edu.umass.cs.mallet.base.classify

Class BalancedWinnow

java.lang.Object

edu.umass.cs.mallet.base.classify.Classifier

edu.umass.cs.mallet.base.classify.BalancedWinnow

All Implemented Interfaces:

Serializable

public class BalancedWinnow
extends Classifier
implements Serializable

Classification methods of BalancedWinnow algorithm.

Author:

Gary Huang ghuang@cs.umass.edu

See Also:

BalancedWinnowTrainer, Serialized Form

Field Summary

Fields inherited from class edu.umass.cs.mallet.base.classify.Classifier

instancePipe

Constructor Summary

Constructors

Constructor and Description
BalancedWinnow(Pipe dataPipe, double[][] weights) Passes along data pipe and weights from BalancedWinnowTrainer

Method Summary

Modifier and Type	Method and Description
Classification	classify(Instance instance) Classifies an instance using BalancedWinnow's weights
double[][]	getWeights()

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

classify, classify, classify, getAccuracy, getAccuracy, getAlphabet, getF1, getF1, getF1, getF1, getInstancePipe, getLabelAlphabet, getPrecision, getPrecision, getPrecision, getPrecision, getRecall, getRecall, getRecall, getRecall, print, print

Methods inherited from class java.lang.Object

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

Constructor Detail

BalancedWinnow

public BalancedWinnow(Pipe dataPipe,
                      double[][] weights)

Passes along data pipe and weights from BalancedWinnowTrainer

Parameters:

dataPipe - needed for dictionary, labels, feature vectors, etc

weights - weights calculated during training phase

Method Detail

getWeights

public double[][] getWeights()

Returns:

a copy of the weight vectors

classify

public Classification classify(Instance instance)

Classifies an instance using BalancedWinnow's weights

Returns a Classification containing the normalized dot products between class weight vectors and the instance feature vector.

One can obtain the confidence of the classification by calculating weight(j')/weight(j), where j' is the highest weight prediction and j is the 2nd-highest. Another possibility is to calculate

e^{dot(w_j', x} / sum_j[e^{dot(w_j, x)}]

Specified by:

classify in class Classifier