org.joone.engine.learning

Class AbstractTeacherSynapse

java.lang.Object

org.joone.engine.Synapse

org.joone.engine.learning.AbstractTeacherSynapse

All Implemented Interfaces:

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

Direct Known Subclasses:

TeacherSynapse

public abstract class AbstractTeacherSynapse
extends Synapse

This class provides a framework to extend in order to implement various teachers, just by overriding or implementing certain functions different functionality can easily implemented.

Author:

Boris Jansen

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected int	currEpoch
protected StreamInputSynapse	desired The stream from where to read the desired input.
protected Fifo	error Into this FIFO (first-in-first-out) object, the calculated error (e.g.
protected boolean	lastErrorPatternReady True if an error pattern is available
protected static ILogger	log Logger
protected NeuralNet	net

Fields inherited from class org.joone.engine.Synapse

array, b_pattern, bitems, bouts, count, enabled, fwdLock, inps, items, learnable, m_batch, m_pattern, myLearner, outs, revLock

Constructor Summary

Constructors

Constructor and Description
AbstractTeacherSynapse() Creates a new instance of AbstractTeacherSynapse
AbstractTeacherSynapse(NeuralNet net)

Method Summary

Methods

Modifier and Type	Method and Description
protected void	backward(double[] pattern) Funzione di TRAIN dell'elemento.
abstract double	calculateError(double aDesired, double anOutput, int anIndex) Calculates the error to be backpropaged for a single output neuron.
abstract double	calculateGlobalError() This method is called after an epoch finished and the global error should be calculated.
java.util.TreeSet	check() Base for check messages.
protected void	constructErrorPattern(double[] aDesired, double[] anOutput) Constructs the error pattern that will be back-propagated.
protected void	epochFinished() This method is called to signal that an epoch has finished.
protected void	forward(double[] pattern) Recall function
Pattern	fwdGet() Here, it forwards (returns) the pushed error (in FIFO order).
void	fwdPut(Pattern pattern) Method to put a pattern forward to the next layer
StreamInputSynapse	getDesired() Gets the stream to read the desired output.
double[]	getLastErrorPattern() Return the most recently seen error pattern
protected int	getSeenPatterns() Get the value of the number of patterns seen during the current epoch.
protected void	incSeenPatterns() Increases the number of seen patterns by one.
void	init() Synapse's initialization.
protected boolean	isFirstTime() Checks whether it is the first time data is passed to this teacher or not.
boolean	isLastErrorPatternReady() See if we have an error pattern available
void	netStoppedError()
protected void	pushError(double error, int count) Pushes the calculated array in the FIFO queue at the end of a epoch, that is after all patterns have been seen.
protected java.lang.Object	readResolve()
void	reset() Reset the input and desired synapses
void	resetInput()
Pattern	revGet() Returns the error pattern coming from the next layer during the training phase
void	revPut(Pattern pattern) Method to put an error pattern backward to the previous layer
protected void	setArrays(int rows, int cols) Insert the method's description here.
boolean	setDesired(StreamInputSynapse newDesired) Set the input data stream containing desired training data.
protected void	setDimensions(int rows, int cols) Dimensiona l'elemento
protected void	setFirstTime(boolean aValue) Sets the first time flag (is it the first time data is forwarded to this teacher).
void	setInputDimension(int newInputDimension) Sets the input dimension of the synapse
void	setMonitor(Monitor newMonitor) Sets the Monitor object of the Teacher Synapse.
protected void	setSeenPatterns(int aValue) Set the value of the number of patterns seen during the current epoch.
protected void	stopTheNet()

Methods inherited from class org.joone.engine.Synapse

addNoise, canCountSteps, fwdPattern_consume, fwdPattern_produce, getCount, getFwdLock, getIgnoreBefore, getInputDimension, getLearner, getLearningRate, getMomentum, getMonitor, getName, getOutputDimension, getRevLock, getSynapseNameSafely, getWeights, initLearner, InspectableTitle, Inspections, isEnabled, isInputFull, isLoopBack, isOutputFull, randomize, revPattern_consume, revPattern_produce, setEnabled, setIgnoreBefore, setInputFull, setLearningRate, setLoopBack, setMomentum, setName, setOutputDimension, setOutputFull, setWeights, warnLogger, warnLogger

Methods inherited from class java.lang.Object

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

Field Detail

log

protected static final ILogger log

Logger

desired

protected StreamInputSynapse desired

The stream from where to read the desired input.

error

protected transient Fifo error

Into this FIFO (first-in-first-out) object, the calculated error (e.g. RMSE) after an epoch will be pushed. This way an (external) application/component is able to read the errors at any moment, providing a loose-coupling mechanism.

currEpoch

protected transient int currEpoch

net

protected NeuralNet net

lastErrorPatternReady

protected boolean lastErrorPatternReady

True if an error pattern is available

Constructor Detail

AbstractTeacherSynapse

public AbstractTeacherSynapse(NeuralNet net)

AbstractTeacherSynapse

public AbstractTeacherSynapse()

Creates a new instance of AbstractTeacherSynapse

Method Detail

setFirstTime

protected void setFirstTime(boolean aValue)

Sets the first time flag (is it the first time data is forwarded to this teacher).

Parameters:

aValue - value for the first time flag.

isFirstTime

protected boolean isFirstTime()

Checks whether it is the first time data is passed to this teacher or not.

Returns:

true if it is the first time data is passed to this teacher, false otherwise.

backward

protected void backward(double[] pattern)

Description copied from class: Synapse

Funzione di TRAIN dell'elemento.

Specified by:

backward in class Synapse

Parameters:

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

pushError

protected void pushError(double error,
             int count)

Pushes the calculated array in the FIFO queue at the end of a epoch, that is after all patterns have been seen.

Parameters:

error, - the calculated error.

count, - the cycle of the calculated error.

stopTheNet

protected void stopTheNet()

getSeenPatterns

protected int getSeenPatterns()

Get the value of the number of patterns seen during the current epoch.

Returns:

the patterns seen during the current epoch.

setSeenPatterns

protected void setSeenPatterns(int aValue)

Set the value of the number of patterns seen during the current epoch.

Parameters:

aValue - the new value for the number of patterns seen during the current epoch.

incSeenPatterns

protected void incSeenPatterns()

Increases the number of seen patterns by one.

fwdGet

public Pattern fwdGet()

Here, it forwards (returns) the pushed error (in FIFO order).

Specified by:

fwdGet in interface InputPatternListener

Overrides:

fwdGet in class Synapse

Returns:

the pattern holding the error of the network. {@link Synapse#fwdGet()

getDesired

public StreamInputSynapse getDesired()

Gets the stream to read the desired output.

Returns:

the desired output stream.

setDesired

public boolean setDesired(StreamInputSynapse newDesired)

Set the input data stream containing desired training data.

Parameters:

newDesired - the stream from where to read the desired output.

readResolve

protected java.lang.Object readResolve()

Overrides:

readResolve in class Synapse

setArrays

protected void setArrays(int rows,
             int cols)

Description copied from class: Synapse

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

Specified by:

setArrays in class Synapse

setDimensions

protected void setDimensions(int rows,
                 int cols)

Description copied from class: Synapse

Dimensiona l'elemento

Specified by:

setDimensions in class Synapse

setInputDimension

public void setInputDimension(int newInputDimension)

Description copied from class: Synapse

Sets the input dimension of the synapse

Specified by:

setInputDimension in interface OutputPatternListener

Overrides:

setInputDimension in class Synapse

Parameters:

newInputDimension - int

reset

public void reset()

Reset the input and desired synapses

Specified by:

reset in interface InputPatternListener

Overrides:

reset in class Synapse

resetInput

public void resetInput()

setMonitor

public void setMonitor(Monitor newMonitor)

Sets the Monitor object of the Teacher Synapse.

Specified by:

setMonitor in interface NeuralElement

Overrides:

setMonitor in class Synapse

Parameters:

newMonitor - neural.engine.Monitor

netStoppedError

public void netStoppedError()

init

public void init()

Description copied from class: Synapse

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

Overrides:

init in class Synapse

check

public java.util.TreeSet check()

Description copied from class: Synapse

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

Specified by:

check in interface NeuralElement

Overrides:

check in class Synapse

Returns:

validation errors.

See Also:

InputPaternListener, OutputPaternListener

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

Overrides:

revPut in class Synapse

Parameters:

pattern - neural.engine.Pattern

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

Overrides:

revGet in class Synapse

Returns:

neural.engine.Pattern

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

Overrides:

fwdPut in class Synapse

Parameters:

pattern - neural.engine.Pattern

forward

protected void forward(double[] pattern)

Description copied from class: Synapse

Recall function

Specified by:

forward in class Synapse

Parameters:

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

constructErrorPattern

protected void constructErrorPattern(double[] aDesired,
                         double[] anOutput)

Constructs the error pattern that will be back-propagated.

Parameters:

aDesired - the desired pattern

anOutput - the actual output pattern

calculateError

public abstract double calculateError(double aDesired,
                    double anOutput,
                    int anIndex)

Calculates the error to be backpropaged for a single output neuron. (The function should also update the global error internally).

Parameters:

aDesired - the desired output

anOutput - the actual output of a single neuron

anIndex - the index of the output neuron

Returns:

the error to be back propagated

calculateGlobalError

public abstract double calculateGlobalError()

This method is called after an epoch finished and the global error should be calculated.

Returns:

the global error (at the end of an epoch).

epochFinished

protected void epochFinished()

This method is called to signal that an epoch has finished. Better to say is that a new epoch has started, because this method is called when the first pattern of a new epoch arrives at the teacher. New implementations of teachers can overwrite this method for their own use. (Please do call super.epochFinished()).

getLastErrorPattern

public double[] getLastErrorPattern()

Return the most recently seen error pattern

isLastErrorPatternReady

public boolean isLastErrorPatternReady()

See if we have an error pattern available