org.simbrain.network.trainers

Class Trainer

java.lang.Object

org.simbrain.network.trainers.Trainer

Direct Known Subclasses:

CompetitiveTrainer, HopfieldTrainer, IterableTrainer, LMSOffline, SOMTrainer

public abstract class Trainer
extends java.lang.Object

Superclass for all trainer classes, which trains a trainable object, typically a network.

Author:

jeffyoshimi

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	Trainer.DataNotInitializedException Exception thrown when a training algorithm is applied but no data have been initialized.

Field Summary

Fields

Modifier and Type	Field and Description
protected Trainable	network The trainable object to be trained.

Constructor Summary

Constructors

Constructor and Description
Trainer(Trainable network) Construct the trainer and pass in a reference to the trainable element.

Method Summary

Modifier and Type	Method and Description
void	addListener(TrainerListener trainerListener) Add a trainer listener.
abstract void	apply() Apply the algorithm.
void	fireProgressUpdate(java.lang.String progressUpdate, int percentComplete) Notify listeners of an update in training progress.
void	fireTrainingBegin() Notify listeners that training has begin.
void	fireTrainingEnd() Notify listeners that training has ended.
java.util.List<TrainerListener>	getListeners()
static Trainable	getTrainable(Group trainedGroup, java.util.List<Neuron> inputNeurons, java.util.List<Neuron> outputNeurons, double[][] inputData, double[][] targetData) Utility method for creating a trainable object.
Trainable	getTrainableNetwork()
void	removeListener(TrainerListener trainerListener) Remove a trainer listener.
void	revalidateSynapseGroups() Helper function to update synapse groups whose synapses may have changed based on training.

Methods inherited from class java.lang.Object

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

Field Detail

network

protected final Trainable network

The trainable object to be trained.

Constructor Detail

Trainer

public Trainer(Trainable network)

Construct the trainer and pass in a reference to the trainable element.

Parameters:

network - the network to be trained

Method Detail

apply

public abstract void apply()
                    throws Trainer.DataNotInitializedException

Apply the algorithm.

Throws:

Trainer.DataNotInitializedException - when input or target data have not been set.

addListener

public void addListener(TrainerListener trainerListener)

Add a trainer listener.

Parameters:

trainerListener - the listener to add

removeListener

public void removeListener(TrainerListener trainerListener)

Remove a trainer listener.

Parameters:

trainerListener - the listener to add

getListeners

public java.util.List<TrainerListener> getListeners()

Returns:

the listeners

fireTrainingBegin

public void fireTrainingBegin()

Notify listeners that training has begin.

fireTrainingEnd

public void fireTrainingEnd()

Notify listeners that training has ended.

fireProgressUpdate

public void fireProgressUpdate(java.lang.String progressUpdate,
                               int percentComplete)

Notify listeners of an update in training progress. Used by GUI progress bars.

Parameters:

progressUpdate - string description of current state

percentComplete - how far along the training is.

getTrainableNetwork

public Trainable getTrainableNetwork()

Returns:

the network

revalidateSynapseGroups

public void revalidateSynapseGroups()

Helper function to update synapse groups whose synapses may have changed based on training.

getTrainable

public static Trainable getTrainable(Group trainedGroup,
                                     java.util.List<Neuron> inputNeurons,
                                     java.util.List<Neuron> outputNeurons,
                                     double[][] inputData,
                                     double[][] targetData)

Utility method for creating a trainable object.

Parameters:

trainedGroup - the subnet or synapse group being trained.

inputNeurons - the input neurons

outputNeurons - the output neurons

inputData - the input data

targetData - the target data

Returns:

the trainable object