org.neuroph.nnet.learning

Class SimulatedAnnealingLearning

java.lang.Object

org.neuroph.core.learning.LearningRule

org.neuroph.core.learning.IterativeLearning

org.neuroph.core.learning.SupervisedLearning

org.neuroph.nnet.learning.SimulatedAnnealingLearning

All Implemented Interfaces:

Serializable

public class SimulatedAnnealingLearning
extends SupervisedLearning

This class implements a simulated annealing learning rule for supervised neural networks. It is based on the generic SimulatedAnnealing class. It is used in the same manner as any other training class that implements the SupervisedLearning interface.

Simulated annealing is a common training method. It is often used in conjunction with a propagation training method. Simulated annealing can be very good when propagation training has reached a local minimum.

The name and inspiration come from annealing in metallurgy, a technique involving heating and controlled cooling of a material to increase the size of its crystals and reduce their defects. The heat causes the atoms to become unstuck from their initial positions (a local minimum of the internal energy) and wander randomly through states of higher energy; the slow cooling gives them more chances of finding configurations with lower internal energy than the initial one.

Author:

Jeff Heaton (http://www.jeffheaton.com)

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected double	temperature The current temperature.

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

maxError, previousEpochError

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
SimulatedAnnealingLearning(NeuralNetwork network)
SimulatedAnnealingLearning(NeuralNetwork network, double startTemp, double stopTemp, int cycles) Construct a simulated annleaing trainer for a feedforward neural network.

Method Summary

Modifier and Type	Method and Description
protected void	calculateWeightChanges(double[] patternError) Not used.
void	doLearningEpoch(DataSet trainingSet) Perform one simulated annealing epoch.
void	doLearningEpoch(DataSet trainingSet, double randomChance)
NeuralNetwork	getNetwork() Get the best network from the training.
void	randomize(double randomChance) Randomize the weights and thresholds.

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

afterEpoch, beforeEpoch, doBatchWeightsUpdate, getErrorFunction, getMaxError, getMinErrorChange, getMinErrorChangeIterationsCount, getMinErrorChangeIterationsLimit, getPreviousEpochError, getTotalNetworkError, isBatchMode, learn, learn, learnPattern, onStart, setBatchMode, setErrorFunction, setMaxError, setMinErrorChange, setMinErrorChangeIterationsLimit

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

temperature

protected double temperature

The current temperature.

Constructor Detail

SimulatedAnnealingLearning

public SimulatedAnnealingLearning(NeuralNetwork network,
                                  double startTemp,
                                  double stopTemp,
                                  int cycles)

Construct a simulated annleaing trainer for a feedforward neural network.

Parameters:

network - The neural network to be trained.

startTemp - The starting temperature.

stopTemp - The ending temperature.

cycles - The number of cycles in a training iteration.

SimulatedAnnealingLearning

public SimulatedAnnealingLearning(NeuralNetwork network)

Method Detail

getNetwork

public NeuralNetwork getNetwork()

Get the best network from the training.

Returns:

The best network.

randomize

public void randomize(double randomChance)

Randomize the weights and thresholds. This function does most of the work of the class. Each call to this class will randomize the data according to the current temperature. The higher the temperature the more randomness.

Parameters:

randomChance -

doLearningEpoch

public void doLearningEpoch(DataSet trainingSet)

Perform one simulated annealing epoch.

Overrides:

doLearningEpoch in class SupervisedLearning

Parameters:

trainingSet - training set for training network

doLearningEpoch

public void doLearningEpoch(DataSet trainingSet,
                            double randomChance)

calculateWeightChanges

protected void calculateWeightChanges(double[] patternError)

Not used.

Specified by:

calculateWeightChanges in class SupervisedLearning

Parameters:

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