org.encog.engine.network.flat
Class FlatNetwork

java.lang.Object
  org.encog.engine.network.flat.FlatNetwork

All Implemented Interfaces:

Serializable, EngineMachineLearning, EngineNeuralNetwork

Direct Known Subclasses:

FlatNetworkRBF

public class FlatNetwork
extends Object
implements EngineNeuralNetwork, Serializable

Implements a flat (vector based) neural network in the Encog Engine. This is meant to be a very highly efficient feedforward, or simple recurrent, neural network. It uses a minimum of objects and is designed with one principal in mind-- SPEED. Readability, code reuse, object oriented programming are all secondary in consideration. Vector based neural networks are also very good for GPU processing. The flat network classes will make use of the GPU if you have enabled GPU processing. See the Encog class for more info.

See Also:

Serialized Form

Field Summary

static double	DEFAULT_BIAS_ACTIVATION The default bias activation.
static double	NO_BIAS_ACTIVATION The value that indicates that there is no bias activation.

Constructor Summary

FlatNetwork()
Default constructor.

FlatNetwork(FlatLayer[] layers)
Create a flat network from an array of layers.

FlatNetwork(int input, int hidden1, int hidden2, int output, boolean tanh)
Construct a flat neural network.

Method Summary

double	calculateError(EngineIndexableSet data) Calculate the error for this neural network.
void	clearConnectionLimit()
void	clearContext() Clear any context neurons.
FlatNetwork	clone() Clone the network.
void	cloneFlatNetwork(FlatNetwork result)
void	compute(double[] input, double[] output) Calculate the output for the given input.
protected void	computeLayer(int currentLayer) Calculate a layer.
void	decodeNetwork(double[] data) Decode the specified data into the weights of the neural network.
double[]	encodeNetwork() Encode the neural network to an array of doubles.
ActivationFunction[]	getActivationFunctions()
int	getBeginTraining()
double	getConnectionLimit()
int[]	getContextTargetOffset()
int[]	getContextTargetSize()
int	getEncodeLength()
int	getEndTraining()
int	getInputCount()
int[]	getLayerCounts()
int[]	getLayerFeedCounts()
int[]	getLayerIndex()
double[]	getLayerOutput()
int	getNeuronCount()
int	getOutputCount()
int[]	getWeightIndex()
double[]	getWeights()
Class<?>	hasSameActivationFunction() Neural networks with only one type of activation function offer certain optimization options.
void	init(FlatLayer[] layers) Construct a flat network.
boolean	isLimited()
void	randomize() Perform a simple randomization of the weights of the neural network between -1 and 1.
void	randomize(double hi, double lo) Perform a simple randomization of the weights of the neural network between the specified hi and lo.
void	setBeginTraining(int beginTraining)
void	setConnectionLimit(double connectionLimit)
void	setEndTraining(int endTraining)

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_BIAS_ACTIVATION

public static final double DEFAULT_BIAS_ACTIVATION

The default bias activation.

See Also:

Constant Field Values

NO_BIAS_ACTIVATION

public static final double NO_BIAS_ACTIVATION

The value that indicates that there is no bias activation.

See Also:

Constant Field Values

Constructor Detail

FlatNetwork

public FlatNetwork()

Default constructor.

FlatNetwork

public FlatNetwork(FlatLayer[] layers)

Create a flat network from an array of layers.

Parameters:

layers - The layers.

FlatNetwork

public FlatNetwork(int input,
                   int hidden1,
                   int hidden2,
                   int output,
                   boolean tanh)

Construct a flat neural network.

Parameters:

input - Neurons in the input layer.

hidden1 - Neurons in the first hidden layer. Zero for no first hidden layer.

hidden2 - Neurons in the second hidden layer. Zero for no second hidden layer.

output - Neurons in the output layer.

tanh - True if this is a tanh activation, false for sigmoid.

Method Detail

calculateError

public double calculateError(EngineIndexableSet data)

Calculate the error for this neural network. The error is calculated using root-mean-square(RMS).

Parameters:

data - The training set.

Returns:

The error percentage.

clearContext

public void clearContext()

Clear any context neurons.

clone

public FlatNetwork clone()

Clone the network.

Overrides:

clone in class Object

Returns:

A clone of the network.

cloneFlatNetwork

public void cloneFlatNetwork(FlatNetwork result)

compute

public void compute(double[] input,
                    double[] output)

Calculate the output for the given input.

Specified by:

compute in interface EngineMachineLearning

Parameters:

input - The input.

output - Output will be placed here.

computeLayer

protected void computeLayer(int currentLayer)

Calculate a layer.

Parameters:

currentLayer - The layer to calculate.

decodeNetwork

public void decodeNetwork(double[] data)

Decode the specified data into the weights of the neural network. This method performs the opposite of encodeNetwork.

Specified by:

decodeNetwork in interface EngineNeuralNetwork

Parameters:

data - The data to be decoded.

encodeNetwork

public double[] encodeNetwork()

Encode the neural network to an array of doubles. This includes the network weights. To read this into a neural network, use the decodeNetwork method.

Specified by:

encodeNetwork in interface EngineNeuralNetwork

Returns:

The encoded network.

getContextTargetOffset

public int[] getContextTargetOffset()

Returns:

The offset of the context target for each layer.

getContextTargetSize

public int[] getContextTargetSize()

Returns:

The context target size for each layer. Zero if the layer does not feed a context layer.

getEncodeLength

public int getEncodeLength()

Specified by:

getEncodeLength in interface EngineNeuralNetwork

Returns:

The length of the array the network would encode to.

getInputCount

public int getInputCount()

Specified by:

getInputCount in interface EngineMachineLearning

Returns:

The number of input neurons.

getLayerCounts

public int[] getLayerCounts()

Returns:

The number of neurons in each layer.

getLayerFeedCounts

public int[] getLayerFeedCounts()

Returns:

The number of neurons in each layer that are fed by the previous layer.

getLayerIndex

public int[] getLayerIndex()

Returns:

Indexes into the weights for the start of each layer.

getLayerOutput

public double[] getLayerOutput()

Returns:

The output for each layer.

getNeuronCount

public int getNeuronCount()

Returns:

The neuron count.

getOutputCount

public int getOutputCount()

Specified by:

getOutputCount in interface EngineMachineLearning

Returns:

The number of output neurons.

getWeightIndex

public int[] getWeightIndex()

Returns:

The index of each layer in the weight and threshold array.

getWeights

public double[] getWeights()

Returns:

The index of each layer in the weight and threshold array.

hasSameActivationFunction

public Class<?> hasSameActivationFunction()

Neural networks with only one type of activation function offer certain optimization options. This method determines if only a single activation function is used.

Returns:

The number of the single activation function, or -1 if there are no activation functions or more than one type of activation function.

init

public void init(FlatLayer[] layers)

Construct a flat network.

Parameters:

layers - The layers of the network to create.

randomize

public void randomize()

Perform a simple randomization of the weights of the neural network between -1 and 1.

randomize

public void randomize(double hi,
                      double lo)

Perform a simple randomization of the weights of the neural network between the specified hi and lo.

Parameters:

hi - The network high.

lo - The network low.

getBeginTraining

public int getBeginTraining()

Returns:

the beginTraining

setBeginTraining

public void setBeginTraining(int beginTraining)

Parameters:

beginTraining - the beginTraining to set

getEndTraining

public int getEndTraining()

Returns:

the endTraining

setEndTraining

public void setEndTraining(int endTraining)

Parameters:

endTraining - the endTraining to set

getConnectionLimit

public double getConnectionLimit()

Returns:

the connectionLimit

setConnectionLimit

public void setConnectionLimit(double connectionLimit)

Parameters:

connectionLimit - the connectionLimit to set

isLimited

public boolean isLimited()

Returns:

the isLimited

clearConnectionLimit

public void clearConnectionLimit()

getActivationFunctions

public ActivationFunction[] getActivationFunctions()

Returns:

The activation functions.