org.joone.engine

Class Synapse

java.lang.Object

org.joone.engine.Synapse

All Implemented Interfaces:

java.io.Serializable, InputPatternListener, Learnable, LearnableSynapse, NeuralElement, OutputPatternListener, Inspectable

Direct Known Subclasses:

AbstractTeacherSynapse, BufferedSynapse, ComparisonSynapse, DelaySynapse, DirectSynapse, FullSynapse, PatternForwardedSynapse, RbfInputSynapse, StreamInputSynapse, StreamOutputSynapse

public abstract class Synapse
extends java.lang.Object
implements InputPatternListener, OutputPatternListener, LearnableSynapse, java.io.Serializable, Inspectable

The Synapse is the connection element between two Layer objects. Its connections are represented by weights that transport the patterns from a layer to another. These weights are modified in the learning cycles, and represent the 'memory' of the trained neural net. MG: Changed to cater for recurrency.

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected Matrix	array
protected Pattern	b_pattern
protected int	bitems
protected double[]	bouts
protected int	count
protected boolean	enabled
protected java.lang.Object	fwdLock
protected double[]	inps
protected int	items
protected boolean	learnable Contains true if for the current Synapse must be used a Learner instead of a built-in learning algorithm.
protected int	m_batch
protected Pattern	m_pattern
protected Learner	myLearner
protected double[]	outs
protected java.lang.Object	revLock

Constructor Summary

Constructors

Constructor and Description
Synapse() The constructor

Method Summary

Methods

Modifier and Type	Method and Description
void	addNoise(double amplitude) Adds a uniformly distributed noise to the weights of the synapse
protected abstract void	backward(double[] pattern) Funzione di TRAIN dell'elemento.
boolean	canCountSteps() Returns TRUE if the synapse calls the method nextStep() on the Monitor object when the fwdGet() method is called
java.util.TreeSet	check() Base for check messages.
protected abstract void	forward(double[] pattern) Recall function
Pattern	fwdGet() Returns the pattern coming from the previous layer during the recall phase
protected void	fwdPattern_consume(Pattern pattern)
protected Pattern	fwdPattern_produce()
void	fwdPut(Pattern pattern) Method to put a pattern forward to the next layer
int	getCount() Return the current pattern count
protected java.lang.Object	getFwdLock() Getter for property fwdLock.
int	getIgnoreBefore() Returns the number of the ignored cycles at beginning of each epoch.
int	getInputDimension() Returns the input dimension of the synapse.
Learner	getLearner() Returns the appropriate Learner object for this class depending on the Monitor.learningMode property value
double	getLearningRate() Returns the value of the learning rate
double	getMomentum() Returns the value of the momentum
Monitor	getMonitor() Returns the Monitor object attached to the synapse
java.lang.String	getName() Returns the name of the synapse
int	getOutputDimension() Returns the output dimension of the synapse.
protected java.lang.Object	getRevLock() Getter for property revLock.
protected java.lang.String	getSynapseNameSafely()
Matrix	getWeights() Getter for the internal matrix of weights
void	init() Synapse's initialization.
void	initLearner() Initialize the Learner object
java.lang.String	InspectableTitle() Method to get the title to show in the InspectionFrame tab.
java.util.Collection	Inspections() Method to get a collection of inspectable objects.
boolean	isEnabled() Getter for property enabled.
boolean	isInputFull() Getter for property inputFull.
boolean	isLoopBack() Getter for property loopBack.
boolean	isOutputFull() Getter for property outputFull.
void	randomize(double amplitude) Initializes all the weigths of the synapses with random values If RandomWeightInitializer is set (default situation) weights are initialized with uniformly distributed random values.
protected java.lang.Object	readResolve()
void	reset() Resets the internal state to be ready for the next run
Pattern	revGet() Returns the error pattern coming from the next layer during the training phase
protected Pattern	revPattern_consume()
protected void	revPattern_produce(Pattern pattern)
void	revPut(Pattern pattern) Method to put an error pattern backward to the previous layer
protected abstract void	setArrays(int rows, int cols) Insert the method's description here.
protected abstract void	setDimensions(int rows, int cols) Dimensiona l'elemento
void	setEnabled(boolean enabled) Setter for property enabled.
void	setIgnoreBefore(int newIgnoreBefore) Sets the number of the ignored cycles at beginning of each epoch.
void	setInputDimension(int newInputDimension) Sets the input dimension of the synapse
void	setInputFull(boolean inputFull) Setter for property inputFull.
void	setLearningRate(double newLearningRate) Deprecated.
void	setLoopBack(boolean loopBack) Setter for property loopBack.
void	setMomentum(double newMomentum) Deprecated.
void	setMonitor(Monitor newMonitor) Sets the Monitor object of the synapse
void	setName(java.lang.String name) Sets the name of the synapse
void	setOutputDimension(int newOutputDimension) Sets the output dimension of the synapse
void	setOutputFull(boolean outputFull) Setter for property outputFull.
void	setWeights(Matrix newWeights) Setter for the internal matrix of weights
protected void	warnLogger(java.lang.String msg, java.lang.Exception ex, ILogger logger)
protected void	warnLogger(java.lang.String msg, ILogger logger)

Methods inherited from class java.lang.Object

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

Field Detail

array

protected Matrix array

m_batch

protected int m_batch

enabled

protected boolean enabled

inps

protected transient double[] inps

outs

protected transient double[] outs

bouts

protected transient double[] bouts

items

protected transient int items

bitems

protected transient int bitems

m_pattern

protected transient Pattern m_pattern

b_pattern

protected transient Pattern b_pattern

count

protected transient int count

myLearner

protected transient Learner myLearner

fwdLock

protected transient volatile java.lang.Object fwdLock

revLock

protected transient volatile java.lang.Object revLock

learnable

protected boolean learnable

Contains true if for the current Synapse must be used a Learner instead of a built-in learning algorithm. Set it in the constructor of any inherited class. Used by the getLearner method.

See Also:

getLearner

Constructor Detail

Synapse

public Synapse()

The constructor

Method Detail

addNoise

public void addNoise(double amplitude)

Adds a uniformly distributed noise to the weights of the synapse

Parameters:

amplitude - Amplitude of the noise: the value is centered around the zero. e.g.: an amplitude = 0.2 means a noise range from -0.2 to 0.2

randomize

public void randomize(double amplitude)

Initializes all the weigths of the synapses with random values If RandomWeightInitializer is set (default situation) weights are initialized with uniformly distributed random values. Otherwise amplitude parameter is ignored and WeightInitializer.initialize() is executed.

Parameters:

amplitude - Amplitude of the random values: the value is centered around the zero. e.g.: an amplitude = 0.2 means a values' range from -0.2 to 0.2

backward

protected abstract void backward(double[] pattern)

Funzione di TRAIN dell'elemento.

Parameters:

pattern - double[] - pattern di input sul quale applicare la funzione di trasferimento

canCountSteps

public boolean canCountSteps()

Returns TRUE if the synapse calls the method nextStep() on the Monitor object when the fwdGet() method is called

Returns:

boolean

forward

protected abstract void forward(double[] pattern)

Recall function

Parameters:

pattern - double[] - input pattern. THIS PATTERN CANNOT BE CHANGED

fwdGet

public Pattern fwdGet()

Description copied from interface: InputPatternListener

Returns the pattern coming from the previous layer during the recall phase

Specified by:

fwdGet in interface InputPatternListener

Returns:

neural.engine.Pattern

fwdPattern_produce

protected Pattern fwdPattern_produce()

fwdPut

public void fwdPut(Pattern pattern)

Description copied from interface: OutputPatternListener

Method to put a pattern forward to the next layer

Specified by:

fwdPut in interface OutputPatternListener

Parameters:

pattern - neural.engine.Pattern

fwdPattern_consume

protected void fwdPattern_consume(Pattern pattern)

Parameters:

pattern -

reset

public void reset()

Resets the internal state to be ready for the next run

Specified by:

reset in interface InputPatternListener

getIgnoreBefore

public int getIgnoreBefore()

Returns the number of the ignored cycles at beginning of each epoch. During these cycles the synapse returns null on the call to the xxxGet methods

Returns:

int

See Also:

setIgnoreBefore(int)

getInputDimension

public int getInputDimension()

Returns the input dimension of the synapse.

Specified by:

getInputDimension in interface LearnableSynapse

Specified by:

getInputDimension in interface OutputPatternListener

Returns:

int

getLearningRate

public double getLearningRate()

Returns the value of the learning rate

Specified by:

getLearningRate in interface LearnableSynapse

Returns:

double

getMomentum

public double getMomentum()

Returns the value of the momentum

Specified by:

getMomentum in interface LearnableSynapse

Returns:

double

getMonitor

public Monitor getMonitor()

Returns the Monitor object attached to the synapse

Specified by:

getMonitor in interface Learnable

Specified by:

getMonitor in interface NeuralElement

Returns:

neural.engine.Monitor

getName

public java.lang.String getName()

Returns the name of the synapse

Specified by:

getName in interface NeuralElement

Returns:

String

See Also:

setName(java.lang.String)

getOutputDimension

public int getOutputDimension()

Returns the output dimension of the synapse.

Specified by:

getOutputDimension in interface InputPatternListener

Specified by:

getOutputDimension in interface LearnableSynapse

Returns:

int

readResolve

protected java.lang.Object readResolve()

revGet

public Pattern revGet()

Description copied from interface: OutputPatternListener

Returns the error pattern coming from the next layer during the training phase

Specified by:

revGet in interface OutputPatternListener

Returns:

neural.engine.Pattern

revPattern_consume

protected Pattern revPattern_consume()

revPut

public void revPut(Pattern pattern)

Description copied from interface: InputPatternListener

Method to put an error pattern backward to the previous layer

Specified by:

revPut in interface InputPatternListener

Parameters:

pattern - neural.engine.Pattern

revPattern_produce

protected void revPattern_produce(Pattern pattern)

Parameters:

pattern -

setArrays

protected abstract void setArrays(int rows,
             int cols)

Insert the method's description here. Creation date: (23/09/2000 12.52.58)

setDimensions

protected abstract void setDimensions(int rows,
                 int cols)

Dimensiona l'elemento

Parameters:

int - rows - righe

int - cols - colonne

setIgnoreBefore

public void setIgnoreBefore(int newIgnoreBefore)

Sets the number of the ignored cycles at beginning of each epoch. During these cycles the synapse is disabled. Useful when the synapse is attached as the Input2 of a SwitchSynapse

Parameters:

newIgnoreBefore - int

See Also:

SwitchSynapse

setInputDimension

public void setInputDimension(int newInputDimension)

Sets the input dimension of the synapse

Specified by:

setInputDimension in interface OutputPatternListener

Parameters:

newInputDimension - int

setLearningRate

public void setLearningRate(double newLearningRate)

Deprecated.

Sets the value of the learning rate Used by library. This value is reseted every teach cycle and is never read by teachers

Parameters:

newLearningRate - double

setMomentum

public void setMomentum(double newMomentum)

Deprecated.

Sets the value of the momentum rate Used by library. This value is reseted every teach cycle.

Parameters:

newMomentum - double

setMonitor

public void setMonitor(Monitor newMonitor)

Sets the Monitor object of the synapse

Specified by:

setMonitor in interface NeuralElement

Parameters:

newMonitor - neural.engine.Monitor

setName

public void setName(java.lang.String name)

Sets the name of the synapse

Specified by:

setName in interface NeuralElement

Parameters:

name - The name of the component.

See Also:

getName()

setOutputDimension

public void setOutputDimension(int newOutputDimension)

Sets the output dimension of the synapse

Specified by:

setOutputDimension in interface InputPatternListener

Parameters:

newOutputDimension - int

isEnabled

public boolean isEnabled()

Getter for property enabled.

Specified by:

isEnabled in interface NeuralElement

Returns:

Value of property enabled.

setEnabled

public void setEnabled(boolean enabled)

Setter for property enabled.

Specified by:

setEnabled in interface NeuralElement

Parameters:

enabled - New value of property enabled.

isLoopBack

public boolean isLoopBack()

Getter for property loopBack.

Returns:

Value of property loopBack.

setLoopBack

public void setLoopBack(boolean loopBack)

Setter for property loopBack.

Parameters:

loopBack - New value of property loopBack.

check

public java.util.TreeSet check()

Base for check messages. Subclasses should call this method from thier own check method.

Specified by:

check in interface NeuralElement

Returns:

validation errors.

See Also:

InputPaternListener, OutputPaternListener

Inspections

public java.util.Collection Inspections()

Description copied from interface: Inspectable

Method to get a collection of inspectable objects.

Specified by:

Inspections in interface Inspectable

Returns:

list of Inspectable objects

See Also:

org.joone.Inspection

InspectableTitle

public java.lang.String InspectableTitle()

Description copied from interface: Inspectable

Method to get the title to show in the InspectionFrame tab.

Specified by:

InspectableTitle in interface Inspectable

Returns:

title of the class.

See Also:

org.joone.InspectionFrame

isInputFull

public boolean isInputFull()

Getter for property inputFull.

Specified by:

isInputFull in interface InputPatternListener

Returns:

Value of property inputFull.

setInputFull

public void setInputFull(boolean inputFull)

Setter for property inputFull.

Specified by:

setInputFull in interface InputPatternListener

Parameters:

inputFull - New value of property inputFull.

isOutputFull

public boolean isOutputFull()

Getter for property outputFull.

Specified by:

isOutputFull in interface OutputPatternListener

Returns:

Value of property outputFull.

setOutputFull

public void setOutputFull(boolean outputFull)

Setter for property outputFull.

Specified by:

setOutputFull in interface OutputPatternListener

Parameters:

outputFull - New value of property outputFull.

getWeights

public Matrix getWeights()

Getter for the internal matrix of weights

Specified by:

getWeights in interface LearnableSynapse

Returns:

the Matrix containing the 2D array of weights

setWeights

public void setWeights(Matrix newWeights)

Setter for the internal matrix of weights

Specified by:

setWeights in interface LearnableSynapse

Parameters:

the - Matrix containing the 2D array of weights

getLearner

public Learner getLearner()

Returns the appropriate Learner object for this class depending on the Monitor.learningMode property value

Specified by:

getLearner in interface Learnable

Returns:

the Learner object if applicable, otherwise null

See Also:

Learnable.getLearner()

initLearner

public void initLearner()

Initialize the Learner object

Specified by:

initLearner in interface Learnable

See Also:

Learnable.initLearner()

getFwdLock

protected final java.lang.Object getFwdLock()

Getter for property fwdLock.

Returns:

Value of property fwdLock.

getRevLock

protected final java.lang.Object getRevLock()

Getter for property revLock.

Returns:

Value of property revLock.

init

public void init()

Synapse's initialization. It needs to be invoked at the starting of the neural network It's called within the Layer.init() method

Specified by:

init in interface NeuralElement

getSynapseNameSafely

protected java.lang.String getSynapseNameSafely()

Returns:

warnLogger

protected void warnLogger(java.lang.String msg,
              ILogger logger)

warnLogger

protected void warnLogger(java.lang.String msg,
              java.lang.Exception ex,
              ILogger logger)

getCount

public int getCount()

Return the current pattern count