de.htwdd.rosenkoenig.neuro.net
Class InnerLayer

java.lang.Object
  de.htwdd.rosenkoenig.neuro.net.Layer
      de.htwdd.rosenkoenig.neuro.net.InnerLayer

public class InnerLayer
extends Layer

This class represents all hidden layers. The name is a little misleading because due to historical reasons the output layer is of this type, too.

Field Summary

private double[]	gradient The array containing the gradients of all neurons of this layer.
private Synapse	inputSynapse This is the input synapse (i.e the connection from the previous layer).

Fields inherited from class de.htwdd.rosenkoenig.neuro.net.Layer

activationFunction, biased, nettoInput, neuronsThatAreNotBiasedCount, output, outputSynapse

Constructor Summary

InnerLayer()
The default ctor.

InnerLayer(int numNeurons, boolean biased)
Ctor that creates an InnerLayer with numNeurons that is biased or not.

InnerLayer(int numNeurons, boolean biased, ActivationFunction activationFunction)
Ctor that creates a InnerLayer with numNeurons that is biased or not.

Method Summary

double[]	getGradient() Getter of the property gradient
Synapse	getInputSynapse() Getter of the property inputSynapse
void	initForTraining() Initialize the layer for training.
void	propagate() Propagate a pattern through the net using the output of the previous layer as input.
void	setGradient(double[] gradient) Setter of the property gradient
void	setInputSynapse(Synapse inputSynapse) Setter of the property inputSynapse

Methods inherited from class de.htwdd.rosenkoenig.neuro.net.Layer

getActivationFunction, getNettoInput, getNeuronsThatAreNotBiasedCount, getOutput, getOutputSynapse, getSize, isBiased, setActivationFunction, setBiased, setOutputSynapse

Methods inherited from class java.lang.Object

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

Field Detail

inputSynapse

private Synapse inputSynapse

This is the input synapse (i.e the connection from the previous layer).

gradient

private double[] gradient

The array containing the gradients of all neurons of this layer.

Constructor Detail

InnerLayer

public InnerLayer()

The default ctor. Calls Layer's ctor.

InnerLayer

public InnerLayer(int numNeurons,
                  boolean biased)

Ctor that creates an InnerLayer with numNeurons that is biased or not. The activation function remains the identity function.

Parameters:

numNeurons - the number of neurons for this InnerLayer

biased - whether or not this InnerLayer is a biased one

InnerLayer

public InnerLayer(int numNeurons,
                  boolean biased,
                  ActivationFunction activationFunction)

Ctor that creates a InnerLayer with numNeurons that is biased or not. The activation function remains will be set to the activation function you specify.

Parameters:

numNeurons - the number of neurons for this InnerLayer

biased - whether or not this InnerLayer is a biased one

Method Detail

getInputSynapse

public Synapse getInputSynapse()

Getter of the property inputSynapse

Returns:

Returns the inputSynapse.

setInputSynapse

public void setInputSynapse(Synapse inputSynapse)

Setter of the property inputSynapse

Parameters:

inputSynapse - The inputSynapse to set.

propagate

public void propagate()

Propagate a pattern through the net using the output of the previous layer as input.

Specified by:

propagate in class Layer

getGradient

public double[] getGradient()

Getter of the property gradient

Returns:

Returns the gradients.

setGradient

public void setGradient(double[] gradient)

Setter of the property gradient

Parameters:

gradient - The gradients to set.

initForTraining

public final void initForTraining()

Initialize the layer for training. This implies the initialization of the inputSynapse (clear all weight changes) and the creation of a new gradient array.