org.neuroph.core.learning

Class SupervisedLearning

java.lang.Object

org.neuroph.core.learning.LearningRule

org.neuroph.core.learning.IterativeLearning

org.neuroph.core.learning.SupervisedLearning

All Implemented Interfaces:

Serializable

Direct Known Subclasses:

LMS, SimulatedAnnealingLearning

public abstract class SupervisedLearning
extends IterativeLearning
implements Serializable

Base class for all supervised learning algorithms. It extends IterativeLearning, and provides general supervised learning principles. Based on Template Method Pattern with abstract method calculateWeightChanges TODO: random pattern order

Author:

Zoran Sevarac

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected double	maxError Max allowed network error (condition to stop learning)
protected double	previousEpochError Total network error in previous epoch

Fields inherited from class org.neuroph.core.learning.IterativeLearning

currentIteration, learningRate, stopConditions

Fields inherited from class org.neuroph.core.learning.LearningRule

listeners, neuralNetwork, trainingSet

Constructor Summary

Constructors

Constructor and Description
SupervisedLearning() Creates new supervised learning rule

Method Summary

Modifier and Type	Method and Description
protected void	afterEpoch()
protected void	beforeEpoch()
protected abstract void	calculateWeightChanges(double[] outputError) This method should implement the weights update procedure for the whole network for the given output error vector.
protected void	doBatchWeightsUpdate() This method updates network weights in batch mode - use accumulated weights change stored in Weight.deltaWeight It is executed after each learning epoch, only if learning is done in batch mode.
void	doLearningEpoch(DataSet trainingSet) This method implements basic logic for one learning epoch for the supervised learning algorithms.
ErrorFunction	getErrorFunction()
double	getMaxError() Returns learning error tolerance - the value of total network error to stop learning.
double	getMinErrorChange() Returns min error change stopping criteria
int	getMinErrorChangeIterationsCount() Returns number of iterations count for for min error change stopping criteria
int	getMinErrorChangeIterationsLimit() Returns number of iterations for min error change stopping criteria
double	getPreviousEpochError() Returns total network error in previous learning epoch
double	getTotalNetworkError()
boolean	isBatchMode() Returns true if learning is performed in batch mode, false otherwise
void	learn(DataSet trainingSet, double maxError) Trains network for the specified training set and maxError
void	learn(DataSet trainingSet, double maxError, int maxIterations) Trains network for the specified training set, maxError and number of iterations
protected void	learnPattern(DataSetRow trainingElement) Trains network with the input and desired output pattern from the specified training element
protected void	onStart() This method is executed when learning starts, before the first epoch.
void	setBatchMode(boolean batchMode) Sets batch mode on/off (true/false)
void	setErrorFunction(ErrorFunction errorFunction)
void	setMaxError(double maxError) Sets allowed network error, which indicates when to stopLearning training
void	setMinErrorChange(double minErrorChange) Sets min error change stopping criteria
void	setMinErrorChangeIterationsLimit(int minErrorChangeIterationsLimit) Sets number of iterations for min error change stopping criteria

Methods inherited from class org.neuroph.core.learning.IterativeLearning

doOneLearningIteration, getCurrentIteration, getLearningRate, getMaxIterations, hasReachedStopCondition, isIterationsLimited, isPausedLearning, learn, learn, pause, resume, setLearningRate, setMaxIterations

Methods inherited from class org.neuroph.core.learning.LearningRule

addListener, fireLearningEvent, getNeuralNetwork, getTrainingSet, isStopped, onStop, removeListener, setNeuralNetwork, setTrainingSet, stopLearning

Methods inherited from class java.lang.Object

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

Field Detail

previousEpochError

protected transient double previousEpochError

Total network error in previous epoch

maxError

protected double maxError

Max allowed network error (condition to stop learning)

Constructor Detail

SupervisedLearning

public SupervisedLearning()

Creates new supervised learning rule

Method Detail

calculateWeightChanges

protected abstract void calculateWeightChanges(double[] outputError)

This method should implement the weights update procedure for the whole network for the given output error vector.

Parameters:

outputError - output error vector for some network input (aka. patternError, network error) usually the difference between desired and actual output

learn

public final void learn(DataSet trainingSet,
                        double maxError)

Trains network for the specified training set and maxError

Parameters:

trainingSet - training set to learn

maxError - learning stop condition. If maxError is reached learning stops

learn

public final void learn(DataSet trainingSet,
                        double maxError,
                        int maxIterations)

Trains network for the specified training set, maxError and number of iterations

Parameters:

trainingSet - training set to learn

maxError - learning stop condition. if maxError is reached learning stops

maxIterations - maximum number of learning iterations

onStart

protected void onStart()

Description copied from class: IterativeLearning

This method is executed when learning starts, before the first epoch. Used for initialisation.

Overrides:

onStart in class IterativeLearning

beforeEpoch

protected void beforeEpoch()

Overrides:

beforeEpoch in class IterativeLearning

afterEpoch

protected void afterEpoch()

Overrides:

afterEpoch in class IterativeLearning

doLearningEpoch

public void doLearningEpoch(DataSet trainingSet)

This method implements basic logic for one learning epoch for the supervised learning algorithms. Epoch is the one pass through the training set. This method iterates through the training set and trains network for each element. It also sets flag if conditions to stop learning has been reached: network error below some allowed value, or maximum iteration count

Specified by:

doLearningEpoch in class IterativeLearning

Parameters:

trainingSet - training set for training network

learnPattern

protected final void learnPattern(DataSetRow trainingElement)

Trains network with the input and desired output pattern from the specified training element

Parameters:

trainingElement - supervised training element which contains input and desired output

doBatchWeightsUpdate

protected void doBatchWeightsUpdate()

This method updates network weights in batch mode - use accumulated weights change stored in Weight.deltaWeight It is executed after each learning epoch, only if learning is done in batch mode.

See Also:

doLearningEpoch(org.neuroph.core.data.DataSet)

isBatchMode

public boolean isBatchMode()

Returns true if learning is performed in batch mode, false otherwise

Returns:

true if learning is performed in batch mode, false otherwise

setBatchMode

public void setBatchMode(boolean batchMode)

Sets batch mode on/off (true/false)

Parameters:

batchMode - batch mode setting

setMaxError

public void setMaxError(double maxError)

Sets allowed network error, which indicates when to stopLearning training

Parameters:

maxError - network error

getMaxError

public double getMaxError()

Returns learning error tolerance - the value of total network error to stop learning.

Returns:

learning error tolerance

getPreviousEpochError

public double getPreviousEpochError()

Returns total network error in previous learning epoch

Returns:

total network error in previous learning epoch

getMinErrorChange

public double getMinErrorChange()

Returns min error change stopping criteria

Returns:

min error change stopping criteria

setMinErrorChange

public void setMinErrorChange(double minErrorChange)

Sets min error change stopping criteria

Parameters:

minErrorChange - value for min error change stopping criteria

getMinErrorChangeIterationsLimit

public int getMinErrorChangeIterationsLimit()

Returns number of iterations for min error change stopping criteria

Returns:

number of iterations for min error change stopping criteria

setMinErrorChangeIterationsLimit

public void setMinErrorChangeIterationsLimit(int minErrorChangeIterationsLimit)

Sets number of iterations for min error change stopping criteria

Parameters:

minErrorChangeIterationsLimit - number of iterations for min error change stopping criteria

getMinErrorChangeIterationsCount

public int getMinErrorChangeIterationsCount()

Returns number of iterations count for for min error change stopping criteria

Returns:

number of iterations count for for min error change stopping criteria

getErrorFunction

public ErrorFunction getErrorFunction()

setErrorFunction

public void setErrorFunction(ErrorFunction errorFunction)

getTotalNetworkError

public double getTotalNetworkError()