org.encog.neural.networks.training.propagation.scg

Class ScaledConjugateGradient

java.lang.Object

org.encog.ml.train.BasicTraining

org.encog.neural.networks.training.propagation.Propagation

org.encog.neural.networks.training.propagation.scg.ScaledConjugateGradient

All Implemented Interfaces:

MLTrain, BatchSize, GradientWorkerOwner, Train, MultiThreadable

public class ScaledConjugateGradient
extends Propagation

This is a training class that makes use of scaled conjugate gradient methods. It is a very fast and efficient training algorithm.

Field Summary

Fields

Modifier and Type	Field and Description
protected static double	FIRST_LAMBDA The starting value for lambda.
protected static double	FIRST_SIGMA The starting value for sigma.

Fields inherited from class org.encog.neural.networks.training.propagation.Propagation

dropoutRandomSource, gradients, network

Constructor Summary

Constructors

Constructor and Description
ScaledConjugateGradient(ContainsFlat network, MLDataSet training) Construct a training class.

Method Summary

Modifier and Type	Method and Description
void	calculateGradients() Calculate the gradients.
boolean	canContinue() This training type does not support training continue.
void	initOthers() Unused.
void	iteration() Perform one iteration.
TrainingContinuation	pause() This training type does not support training continue.
void	resume(TrainingContinuation state) This training type does not support training continue.
double	updateWeight(double[] gradients, double[] lastGradient, int index) Update the weights.
double	updateWeight(double[] gradients, double[] lastGradient, int index, double dropoutRate) Update the weights.

Methods inherited from class org.encog.neural.networks.training.propagation.Propagation

finishTraining, finishTraining, fixFlatSpot, getBatchSize, getCurrentFlatNetwork, getDropoutRate, getL1, getL2, getLastGradient, getMethod, getThreadCount, iteration, learn, learnLimited, report, rollIteration, setBatchSize, setDroupoutRate, setErrorFunction, setL1, setL2, setThreadCount

Methods inherited from class org.encog.ml.train.BasicTraining

addStrategy, getError, getImplementationType, getIteration, getStrategies, getTraining, isTrainingDone, postIteration, preIteration, setError, setIteration, setTraining

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.encog.ml.train.MLTrain

addStrategy, getError, getImplementationType, getIteration, getStrategies, getTraining, isTrainingDone, setError, setIteration

Field Detail

FIRST_SIGMA

protected static final double FIRST_SIGMA

The starting value for sigma.

See Also:

Constant Field Values

FIRST_LAMBDA

protected static final double FIRST_LAMBDA

The starting value for lambda.

See Also:

Constant Field Values

Constructor Detail

ScaledConjugateGradient

public ScaledConjugateGradient(ContainsFlat network,
                               MLDataSet training)

Construct a training class.

Parameters:

network - The network to train.

training - The training data.

Method Detail

canContinue

public boolean canContinue()

This training type does not support training continue.

Returns:

Always returns false.

pause

public TrainingContinuation pause()

This training type does not support training continue.

Returns:

Always returns null.

resume

public void resume(TrainingContinuation state)

This training type does not support training continue.

Parameters:

state - Not used.

calculateGradients

public void calculateGradients()

Calculate the gradients. They are normalized as well.

Overrides:

calculateGradients in class Propagation

iteration

public void iteration()

Perform one iteration.

Specified by:

iteration in interface MLTrain

Overrides:

iteration in class Propagation

updateWeight

public double updateWeight(double[] gradients,
                           double[] lastGradient,
                           int index)

Update the weights.

Specified by:

updateWeight in class Propagation

Parameters:

gradients - The current gradients.

lastGradient - The last gradients.

index - The weight index being updated.

Returns:

The new weight value.

updateWeight

public double updateWeight(double[] gradients,
                           double[] lastGradient,
                           int index,
                           double dropoutRate)

Update the weights.

Specified by:

updateWeight in class Propagation

Parameters:

gradients - The current gradients.

lastGradient - The last gradients.

index - The weight index being updated.

dropoutRate - The dropout rate

Returns:

The new weight value.

initOthers

public void initOthers()

Unused.

Specified by:

initOthers in class Propagation