ir.classifiers
Class PerceptronUnit

java.lang.Object
  ir.classifiers.PerceptronUnit

public class PerceptronUnit
extends java.lang.Object

An individual perceptron object used by the Perceptron classifier Includes methods for classifying an Example and training the unit to fire only for a given category of examples.

Field Summary

protected java.lang.String[]	categories Array of categories (classes) in the data
boolean	debug Flag for debug print statements
double	learningRate The learning rate for weight updating
int	maxEpochs Maximum number of training epochs allowed
protected double	threshold The threshold of the perceptron.
protected HashMapVector	weights The weights of the perceptron.

Constructor Summary

PerceptronUnit(java.lang.String[] categories, boolean debug)
Create an PerceptronUnit with these attributes

Method Summary

double	classify(Example example) Classify an example using the perceptron and return the amount by which the net input exceeds the threshold as a measure of confidence of the prediction of being positive for this perceptron
void	clear() Clear the weights and threshold all back to zero
void	trainCategory(java.util.List trainExamples, int category) Trains the perceptron to only fire for examples in the given category, i.e.

Methods inherited from class java.lang.Object

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

Field Detail

threshold

protected double threshold

The threshold of the perceptron. The net input must exceed this threshold in order for the neuron to fire

weights

protected HashMapVector weights

The weights of the perceptron. Since the weight vector is sparse, most entries are zero, a HashMapVector is used to store the weights for the non-zero features (tokens). See ir.vsr.HashMapVector.

maxEpochs

public int maxEpochs

Maximum number of training epochs allowed

learningRate

public double learningRate

The learning rate for weight updating

categories

protected java.lang.String[] categories

Array of categories (classes) in the data

debug

public boolean debug

Flag for debug print statements

Constructor Detail

PerceptronUnit

public PerceptronUnit(java.lang.String[] categories,
                      boolean debug)

Create an PerceptronUnit with these attributes

Parameters:

categories - The array of Strings containing the category names

debug - Flag to turn on detailed output

Method Detail

clear

public void clear()

Clear the weights and threshold all back to zero

classify

public double classify(Example example)

Classify an example using the perceptron and return the amount by which the net input exceeds the threshold as a measure of confidence of the prediction of being positive for this perceptron

trainCategory

public void trainCategory(java.util.List trainExamples,
                          int category)

Trains the perceptron to only fire for examples in the given category, i.e. examples in the given category should be treated as positive examples for this perceptron and examples in all other categories should be treated as negative