org.encog.neural.networks.training.propagation

Class GradientWorker

java.lang.Object

org.encog.neural.networks.training.propagation.GradientWorker

All Implemented Interfaces:

EngineTask

public class GradientWorker
extends java.lang.Object
implements EngineTask

Worker class for the mulithreaded training of flat networks.

Field Summary

Fields

Modifier and Type	Field and Description
protected java.util.Random	dropoutRandomSource Used to generate randomness for dropout

Constructor Summary

Constructors

Constructor and Description
GradientWorker(FlatNetwork theNetwork, GradientWorkerOwner theOwner, MLDataSet theTraining, int theLow, int theHigh, double[] flatSpot, ErrorFunction ef) Construct a gradient worker.

Method Summary

Modifier and Type	Method and Description
void	calculateRegularizationPenalty(double[] l)
ErrorCalculation	getErrorCalculation()
double[]	getGradients()
FlatNetwork	getNetwork()
double[]	getWeights()
void	layerRegularizationPenalty(int fromLayer, double[] l)
void	process(MLDataPair pair) Process one training set element.
void	run() Perform the gradient calculation for the specified index range.
void	run(int index)

Methods inherited from class java.lang.Object

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

Field Detail

dropoutRandomSource

protected java.util.Random dropoutRandomSource

Used to generate randomness for dropout

Constructor Detail

GradientWorker

public GradientWorker(FlatNetwork theNetwork,
                      GradientWorkerOwner theOwner,
                      MLDataSet theTraining,
                      int theLow,
                      int theHigh,
                      double[] flatSpot,
                      ErrorFunction ef)

Construct a gradient worker.

Parameters:

theNetwork - The network to train.

theOwner - The owner that is doing the training.

theTraining - The training data.

theLow - The low index to use in the training data.

theHigh - The high index to use in the training data.

flatSpot - The flatspot additions for each layer

ef - Error function

Method Detail

getNetwork

public FlatNetwork getNetwork()

Returns:

The network being processed.

getWeights

public double[] getWeights()

Returns:

The weights for this network.

process

public void process(MLDataPair pair)

Process one training set element.

Parameters:

pair - the training data information

run

public final void run()

Perform the gradient calculation for the specified index range.

Specified by:

run in interface EngineTask

run

public final void run(int index)

getErrorCalculation

public ErrorCalculation getErrorCalculation()

getGradients

public double[] getGradients()

Returns:

the gradients

calculateRegularizationPenalty

public void calculateRegularizationPenalty(double[] l)

layerRegularizationPenalty

public void layerRegularizationPenalty(int fromLayer,
                                       double[] l)