edu.umass.cs.mallet.base.classify

Class BalancedWinnowTrainer

java.lang.Object

edu.umass.cs.mallet.base.classify.ClassifierTrainer

edu.umass.cs.mallet.base.classify.BalancedWinnowTrainer

All Implemented Interfaces:

Boostable, Serializable

public class BalancedWinnowTrainer
extends ClassifierTrainer
implements Boostable, Serializable

An implementation of the training methods of a BalancedWinnow on-line classifier. Given a labeled instance (x, y) the algorithm computes dot(x, wi), for w1, ... , wc where wi is the weight vector for class i. The instance is classified as class j if the value of dot(x, wj) is the largest among the c dot products.

The weight vectors are updated whenever the the classifier makes a mistake or just barely got the correct answer (highest dot product is within delta percent higher than the second highest). Suppose the classifier guessed j and answer was j'. For each feature i that is present, multiply w_ji by (1-epsilon) and multiply w_j'i by (1+epsilon)

The above procedure is done multiple times to the training examples (default is 5), and epsilon is cut by the cooling rate at each iteration (default is cutting epsilon by half).

Author:

Gary Huang ghuang@cs.umass.edu

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
static double	DEFAULT_COOLING_RATE 0.5
static double	DEFAULT_DELTA 0.1
static double	DEFAULT_EPSILON 0.5
static int	DEFAULT_MAX_ITERATIONS 30

Constructor Summary

Constructors

Constructor and Description
BalancedWinnowTrainer() Default constructor.
BalancedWinnowTrainer(double epsilon, double delta, int maxIterations, double coolingRate)

Method Summary

Modifier and Type	Method and Description
Classifier	train(InstanceList trainingList, InstanceList validationList, InstanceList testSet, ClassifierEvaluating evaluator, Classifier initialClassifier) Trains the classifier on the instance list, updating class weight vectors as appropriate

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

Field Detail

DEFAULT_EPSILON

public static final double DEFAULT_EPSILON

0.5

See Also:

Constant Field Values

DEFAULT_DELTA

public static final double DEFAULT_DELTA

0.1

See Also:

Constant Field Values

DEFAULT_MAX_ITERATIONS

public static final int DEFAULT_MAX_ITERATIONS

30

See Also:

Constant Field Values

DEFAULT_COOLING_RATE

public static final double DEFAULT_COOLING_RATE

0.5

See Also:

Constant Field Values

Constructor Detail

BalancedWinnowTrainer

public BalancedWinnowTrainer()

Default constructor. Sets all features to defaults.

BalancedWinnowTrainer

public BalancedWinnowTrainer(double epsilon,
                             double delta,
                             int maxIterations,
                             double coolingRate)

Parameters:

epsilon - percentage by which to increase/decrease weight vectors when an example is misclassified.

delta - percentage by which the highest (and correct) dot product should exceed the second highest dot product before we consider an example to be correctly classified (margin width) when adjusting weights.

maxIterations - maximum number of times to loop through training examples.

coolingRate - percentage of epsilon to decrease after each iteration

Method Detail

train

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

Trains the classifier on the instance list, updating class weight vectors as appropriate

Specified by:

train in class ClassifierTrainer

Parameters:

trainingList - Instance list to be trained on

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