uchicago.src.sim.adaptation.neural
Class RepastNeuralWrapper

java.lang.Object
  uchicago.src.sim.adaptation.neural.RepastNeuralWrapper

All Implemented Interfaces:

java.io.Serializable, java.util.EventListener, org.joone.engine.NeuralNetListener

public class RepastNeuralWrapper
extends java.lang.Object
implements org.joone.engine.NeuralNetListener, java.io.Serializable

A wrapper around a NeuralNet. This class adds training and retrieval methods that wait until the training/retrieval is finished before returning. Note: if you want serialize the network that this wrapper wraps, use the saveNetToFile and NeuralUtils.loadNetFromFile methods, otherwise you will have multiple wrappers attached to the serialized networks (through the listeners).

Version:

$Revision: 1.4 $ $Date: 2005/08/12 16:13:29 $

Author:

Jerry Vos

See Also:

Serialized Form

Field Summary

protected int	epochsPerIteration The number of times to apply the patterns to the network during the training
protected org.joone.net.NeuralNet	net The net being wrapped
protected boolean	netStopped Whether or not the network is running (training/retrieving).

Constructor Summary

RepastNeuralWrapper()
The default constructor for a network wrapper.

RepastNeuralWrapper(org.joone.net.NeuralNet net)
This creates a basic network wrapper, however this cannot be used until Layers, Synapses, and so forth have been specified.

Method Summary

void	cicleTerminated(org.joone.engine.NeuralNetEvent e)
void	errorChanged(org.joone.engine.NeuralNetEvent e)
int	getEpochsPerIteration()
org.joone.net.NeuralNet	getNet()
void	netStarted(org.joone.engine.NeuralNetEvent e)
void	netStopped(org.joone.engine.NeuralNetEvent e)
void	netStoppedError(org.joone.engine.NeuralNetEvent e, java.lang.String error)
org.joone.engine.Pattern	retrieve(org.joone.engine.InputPatternListener in) This method will retrieve a result Pattern from a network based on the in parameter.
void	saveNetToFile(java.lang.String fileName) Saves the neural network this is wrapping to the specified file.
void	setEpochsPerIteration(int epochsPerIteration)
void	setNet(org.joone.net.NeuralNet net)
void	train(org.joone.engine.InputPatternListener in) Trains a network.

Methods inherited from class java.lang.Object

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

Field Detail

net

protected org.joone.net.NeuralNet net

The net being wrapped

epochsPerIteration

protected int epochsPerIteration

The number of times to apply the patterns to the network during the training

netStopped

protected transient boolean netStopped

Whether or not the network is running (training/retrieving). This doesn't mean that the network's threads have all shut down, just that the network has signaled its start or stop.

Constructor Detail

RepastNeuralWrapper

public RepastNeuralWrapper()

The default constructor for a network wrapper. This creates a basic network, however this cannot be used until Layers, Synapses, and so forth have been specified.
Same as new RepastNeuralWrapper(new NeuralNet());

RepastNeuralWrapper

public RepastNeuralWrapper(org.joone.net.NeuralNet net)

This creates a basic network wrapper, however this cannot be used until Layers, Synapses, and so forth have been specified.

Method Detail

retrieve

public org.joone.engine.Pattern retrieve(org.joone.engine.InputPatternListener in)
                                  throws NeuralException

This method will retrieve a result Pattern from a network based on the in parameter.
This method will remove all outputs and inputs from the network, and then it will add a DirectSynapse as an output, and the in parameter as an input. THESE ARE NOT RESTORED BY THE CALL.

Parameters:

in - The input to the network (best as a DirectSynapse)

Returns:

the Pattern resulting from running the network

Throws:

NeuralException - when there is an error querying the network

train

public void train(org.joone.engine.InputPatternListener in)
           throws NeuralException

Trains a network. The network is expected to be fully built before this method is called.
This method will remove all outputs and inputs from the network, and then it will add the network's teacher as an output, and the in parameter as an input. THESE ARE NOT RESTORED BY THE CALL.
To use this method the network's teacher should have its desired output specified through a getTeacher().setDesired(..) call. Also the in should be setup to deliver at least one pattern to the network (this training only feeds one pattern through).
Make sure before calling this method you specify the learning rate and momentum of the network's Monitor.

Parameters:

in - the input for the network.

Throws:

NeuralException - when there is an error training the network

See Also:

NeuralNet.check(), retrieve(InputPatternListener)

getEpochsPerIteration

public int getEpochsPerIteration()

Returns:

the number of times to feed a pattern through the network

setEpochsPerIteration

public void setEpochsPerIteration(int epochsPerIteration)

Parameters:

epochsPerIteration - the number of times to feed a pattern through the network

netStarted

public void netStarted(org.joone.engine.NeuralNetEvent e)

Specified by:

netStarted in interface org.joone.engine.NeuralNetListener

netStopped

public void netStopped(org.joone.engine.NeuralNetEvent e)

Specified by:

netStopped in interface org.joone.engine.NeuralNetListener

netStoppedError

public void netStoppedError(org.joone.engine.NeuralNetEvent e,
                            java.lang.String error)

Specified by:

netStoppedError in interface org.joone.engine.NeuralNetListener

cicleTerminated

public void cicleTerminated(org.joone.engine.NeuralNetEvent e)

Specified by:

cicleTerminated in interface org.joone.engine.NeuralNetListener

errorChanged

public void errorChanged(org.joone.engine.NeuralNetEvent e)

Specified by:

errorChanged in interface org.joone.engine.NeuralNetListener

getNet

public org.joone.net.NeuralNet getNet()

Returns:

the net this wrapper is wrapping

setNet

public void setNet(org.joone.net.NeuralNet net)

Parameters:

net - the new net to wrap

saveNetToFile

public void saveNetToFile(java.lang.String fileName)
                   throws NeuralException

Saves the neural network this is wrapping to the specified file. This can be loaded through #loadNetFromFile(String).

Parameters:

fileName - the name of the file to save the network to

Throws:

NeuralException - when there is an error saving the network

See Also:

#loadNetFromFile(String), NeuralUtils.saveNetToFile(NeuralNet, String)