Package me.damoebe.architectures.mlp

Klasse FFNetwork

java.lang.Object

me.damoebe.architectures.Network

me.damoebe.architectures.mlp.FFNetwork

Bekannte direkte Unterklassen:

DeepNetwork, EvolutionNetwork

public abstract class FFNetwork
extends Network

The abstract class for FFNeuralNetworks

Feldübersicht

Felder

Modifizierer und Typ	Feld	Beschreibung
protected double	currentLoss	The current loss of the network
protected final List<Layer>	layers	The whole network in a list
protected final double	learningRate
protected double	noise	The configured noise of the network

Konstruktorübersicht

Konstruktoren

Modifizierer	Konstruktor	Beschreibung
protected	FFNetwork(int inputSize, int outputSize, int hiddenLayerSize, int hiddenLayerAmount)	FFNetwork constructor for networks without a learning rate - 0.0 (for example evolution learning)
protected	FFNetwork(int inputSize, int outputSize, int hiddenLayerSize, int hiddenLayerAmount, double learningRate)	The main constructor of a FFNetwork

Methodenübersicht

Modifizierer und Typ	Methode	Beschreibung
abstract FFNetwork	clone()	A method for cloning a FFNetwork object
static Layer	generateLayer(int size, Layer prevLayer)	Generates a new layer for the FFNetwork constructor
int	getHiddenLayerAmount()	A Getter for the amount of hidden-layers in this network
int	getHiddenLayerSize()	A Getter for the size of the middle layers size
int	getInputSize()	A Getter for the size of the first layer's size
List<Layer>	getLayers()	Getter for the layer-list of this network
abstract double	getNetworkLoss()	Gets the network loss
List<Double>	getOutput()	Gets all output activations
int	getOutputSize()	A Getter for the size of the last layer's size
void	insertInput(List<Double> input)	Inserts input activations into the network
void	setNoise(double noise)	Sets the network noise.
abstract void	train(List<Double> input, List<Double> optimalOutput)	The abstract train function
void	updateAllActivations()	Calls the updatedActivation() function for each neuron in the network
void	updateLoss(List<Double> optimalOutput)	Updates the network loss based on a targetActivation list

Von Klasse geerbte Methoden me.damoebe.architectures.Network

loadNetworkFromJson, loadToJsonFile

Von Klasse geerbte Methoden java.lang.Object

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

Felddetails

layers

protected final List<Layer> layers

The whole network in a list

currentLoss

protected double currentLoss

The current loss of the network

noise

protected double noise

The configured noise of the network

learningRate

protected final double learningRate

Konstruktordetails

FFNetwork

protected FFNetwork(int inputSize,
 int outputSize,
 int hiddenLayerSize,
 int hiddenLayerAmount,
 double learningRate)

The main constructor of a FFNetwork

Parameter:

inputSize - The size of the first layer

outputSize - The size of the last layer

hiddenLayerSize - The size of the hiddenlayers

hiddenLayerAmount - The amount of hiddenlayers that should be generated

FFNetwork

protected FFNetwork(int inputSize,
 int outputSize,
 int hiddenLayerSize,
 int hiddenLayerAmount)

FFNetwork constructor for networks without a learning rate - 0.0 (for example evolution learning)

Parameter:

inputSize - The size of the first layer

outputSize - The size of the last layer

hiddenLayerSize - The size of the hiddenlayers

hiddenLayerAmount - The amount of hiddenlayers that should be generated

Methodendetails

clone

public abstract FFNetwork clone()

A method for cloning a FFNetwork object

Setzt außer Kraft:

clone in Klasse Object

Gibt zurück:

the cloned network object

generateLayer

public static Layer generateLayer(int size,
 Layer prevLayer)

Generates a new layer for the FFNetwork constructor

Parameter:

size - The amount of neurons that layer should contain

prevLayer - The previous Layer in the network (null if there is no previous layer)

Gibt zurück:

A new Layer that points to the previous layer

updateAllActivations

public void updateAllActivations()

Calls the updatedActivation() function for each neuron in the network

Angegeben von:

updateAllActivations in Klasse Network

insertInput

public void insertInput(List<Double> input)

Inserts input activations into the network

Angegeben von:

insertInput in Klasse Network

Parameter:

input - A Double ArrayList containing the neuron activations.

updateLoss

public void updateLoss(List<Double> optimalOutput)

Updates the network loss based on a targetActivation list

Parameter:

optimalOutput - The target activations for the current repetition

getOutput

public List<Double> getOutput()

Gets all output activations

Angegeben von:

getOutput in Klasse Network

Gibt zurück:

A Double List containing all output layer activations.

setNoise

public void setNoise(double noise)

Sets the network noise.

Parameter:

noise - The new noise of the network.

getInputSize

public int getInputSize()

A Getter for the size of the first layer's size

Gibt zurück:

The input-layer size

getOutputSize

public int getOutputSize()

A Getter for the size of the last layer's size

Gibt zurück:

The output-layer size

getHiddenLayerSize

public int getHiddenLayerSize()

A Getter for the size of the middle layers size

Gibt zurück:

The size of the hidden layers

getHiddenLayerAmount

public int getHiddenLayerAmount()

A Getter for the amount of hidden-layers in this network

Gibt zurück:

The hidden-layer amount

train

public abstract void train(List<Double> input,
 List<Double> optimalOutput)

The abstract train function

Angegeben von:

train in Klasse Network

Parameter:

input - The inputs for this repetition

optimalOutput - The optimalOutputs for this repetition

getNetworkLoss

public abstract double getNetworkLoss()

Gets the network loss

Gibt zurück:

The network loss double

getLayers

public List<Layer> getLayers()

Getter for the layer-list of this network

Gibt zurück:

The layers in a list