org.encog.neural.freeform

Class FreeformNetwork

java.lang.Object

org.encog.ml.BasicML

org.encog.neural.freeform.FreeformNetwork

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable, MLClassification, MLContext, MLEncodable, MLError, MLInput, MLInputOutput, MLMethod, MLOutput, MLProperties, MLRegression, MLResettable

public class FreeformNetwork
extends BasicML
implements MLContext, java.lang.Cloneable, MLRegression, MLEncodable, MLResettable, MLClassification, MLError

Implements a freefrom neural network. A freeform neural network can represent much more advanced structures than the flat networks that the Encog BasicNetwork implements. However, while freeform networks are more advanced than the BasicNetwork, they are also much slower. Freeform networks allow just about any neuron to be connected to another neuron. You can have neuron layers if you want, but they are not required.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
FreeformNetwork() Default constructor.
FreeformNetwork(BasicNetwork network) Craete a freeform network from a basic network.

Method Summary

Modifier and Type	Method and Description
double	calculateError(MLDataSet data) Calculate the error of the ML method, given a dataset.
int	classify(MLData input) Classify the input into a group.
void	clearContext() Clear the context.
java.lang.Object	clone() Return a clone of this neural network.
MLData	compute(MLData input) Compute regression.
void	connectLayers(FreeformLayer source, FreeformLayer target) Connect two layers.
void	ConnectLayers(FreeformLayer source, FreeformLayer target, ActivationFunction theActivationFunction) Connect two layers, assume bias activation of 1.0 and non-recurrent connection.
void	connectLayers(FreeformLayer source, FreeformLayer target, ActivationFunction theActivationFunction, double biasActivation, boolean isRecurrent) Connect two layers.
FreeformLayer	createContext(FreeformLayer source, FreeformLayer target) Create a context connection, such as those used by Jordan/Elmann.
static FreeformNetwork	createElman(int input, int hidden1, int output, ActivationFunction af) Construct an Elmann recurrent neural network.
static FreeformNetwork	createFeedforward(int input, int hidden1, int hidden2, int output, ActivationFunction af) Create a feedforward freeform neural network.
FreeformLayer	createInputLayer(int neuronCount) Create the input layer.
FreeformLayer	createLayer(int neuronCount) Create a hidden layer.
FreeformLayer	createOutputLayer(int neuronCount) Create the output layer.
void	decodeFromArray(double[] encoded) Decode an array to this object.
int	encodedArrayLength()
void	encodeToArray(double[] encoded) Encode the object to the specified array.
int	getInputCount()
int	getOutputCount()
FreeformLayer	getOutputLayer()
void	performConnectionTask(ConnectionTask task) Perform the specified connection task.
void	performNeuronTask(NeuronTask task) Perform the specified neuron task.
void	reset() Reset the weights.
void	reset(int seed) Reset the weights with a seed.
void	tempTrainingAllocate(int neuronSize, int connectionSize) Allocate temp training space.
void	tempTrainingClear() Clear the temp training data.
void	updateContext() Update context.
void	updateProperties() Update any objeccts when a property changes.

Methods inherited from class org.encog.ml.BasicML

getProperties, getPropertyDouble, getPropertyLong, getPropertyString, setProperty, setProperty, setProperty

Methods inherited from class java.lang.Object

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

Constructor Detail

FreeformNetwork

public FreeformNetwork()

Default constructor. Typically should not be directly used.

FreeformNetwork

public FreeformNetwork(BasicNetwork network)

Craete a freeform network from a basic network.

Parameters:

network - The basic network to use.

Method Detail

createElman

public static FreeformNetwork createElman(int input,
                                          int hidden1,
                                          int output,
                                          ActivationFunction af)

Construct an Elmann recurrent neural network.

Parameters:

input - The input count.

hidden1 - The hidden count.

output - The output count.

af - The activation function.

Returns:

The newly created network.

createFeedforward

public static FreeformNetwork createFeedforward(int input,
                                                int hidden1,
                                                int hidden2,
                                                int output,
                                                ActivationFunction af)

Create a feedforward freeform neural network.

Parameters:

input - The input count.

hidden1 - The first hidden layer count, zero if none.

hidden2 - The second hidden layer count, zero if none.

output - The output count.

af - The activation function.

Returns:

The newly crated network.

calculateError

public double calculateError(MLDataSet data)

Calculate the error of the ML method, given a dataset.

Specified by:

calculateError in interface MLError

Parameters:

data - The dataset.

Returns:

The error.

classify

public int classify(MLData input)

Classify the input into a group.

Specified by:

classify in interface MLClassification

Parameters:

input - The input data to classify.

Returns:

The group that the data was classified into.

clearContext

public void clearContext()

Clear the context.

Specified by:

clearContext in interface MLContext

clone

public java.lang.Object clone()

Return a clone of this neural network. Including structure, weights and bias values. This is a deep copy.

Overrides:

clone in class java.lang.Object

Returns:

A cloned copy of the neural network.

compute

public MLData compute(MLData input)

Compute regression.

Specified by:

compute in interface MLRegression

Parameters:

input - The input data.

Returns:

The output data.

connectLayers

public void connectLayers(FreeformLayer source,
                          FreeformLayer target)

Connect two layers. These layers will be connected with a TANH activation function in a non-recurrent way. A bias activation of 1.0 will be used, if needed.

Parameters:

source - The source layer.

target - The target layer.

connectLayers

public void connectLayers(FreeformLayer source,
                          FreeformLayer target,
                          ActivationFunction theActivationFunction,
                          double biasActivation,
                          boolean isRecurrent)

Connect two layers.

Parameters:

source - The source layer.

target - The target layer.

theActivationFunction - The activation function to use.

biasActivation - The bias activation to use.

isRecurrent - True, if this is a recurrent connection.

ConnectLayers

public void ConnectLayers(FreeformLayer source,
                          FreeformLayer target,
                          ActivationFunction theActivationFunction)

Connect two layers, assume bias activation of 1.0 and non-recurrent connection.

Parameters:

source - The source layer.

target - The target layer.

theActivationFunction - The activation function.

createContext

public FreeformLayer createContext(FreeformLayer source,
                                   FreeformLayer target)

Create a context connection, such as those used by Jordan/Elmann.

Parameters:

source - The source layer.

target - The target layer.

Returns:

The newly created context layer.

createInputLayer

public FreeformLayer createInputLayer(int neuronCount)

Create the input layer.

Parameters:

neuronCount - The input neuron count.

Returns:

The newly created layer.

createLayer

public FreeformLayer createLayer(int neuronCount)

Create a hidden layer.

Parameters:

neuronCount - The neuron count.

Returns:

The newly created layer.

createOutputLayer

public FreeformLayer createOutputLayer(int neuronCount)

Create the output layer.

Parameters:

neuronCount - The neuron count.

Returns:

The newly created output layer.

decodeFromArray

public void decodeFromArray(double[] encoded)

Decode an array to this object.

Specified by:

decodeFromArray in interface MLEncodable

Parameters:

encoded - The encoded array.

encodedArrayLength

public int encodedArrayLength()

Specified by:

encodedArrayLength in interface MLEncodable

Returns:

The length of an encoded array.

encodeToArray

public void encodeToArray(double[] encoded)

Encode the object to the specified array.

Specified by:

encodeToArray in interface MLEncodable

Parameters:

encoded - The array.

getInputCount

public int getInputCount()

Specified by:

getInputCount in interface MLInput

Returns:

The input.

getOutputCount

public int getOutputCount()

Specified by:

getOutputCount in interface MLOutput

Returns:

The output count.

getOutputLayer

public FreeformLayer getOutputLayer()

Returns:

The output layer.

performConnectionTask

public void performConnectionTask(ConnectionTask task)

Perform the specified connection task. This task will be performed over all connections.

Parameters:

task - The connection task.

performNeuronTask

public void performNeuronTask(NeuronTask task)

Perform the specified neuron task. This task will be executed over all neurons.

Parameters:

task - The neuron task to perform.

reset

public void reset()

Reset the weights.

Specified by:

reset in interface MLResettable

reset

public void reset(int seed)

Reset the weights with a seed.

Specified by:

reset in interface MLResettable

Parameters:

seed - The seed value.

tempTrainingAllocate

public void tempTrainingAllocate(int neuronSize,
                                 int connectionSize)

Allocate temp training space.

Parameters:

neuronSize - The number of elements to allocate on each neuron.

connectionSize - The number of elements to allocate on each connection.

tempTrainingClear

public void tempTrainingClear()

Clear the temp training data.

updateContext

public void updateContext()

Update context.

updateProperties

public void updateProperties()

Update any objeccts when a property changes.

Specified by:

updateProperties in interface MLProperties

Specified by:

updateProperties in class BasicML