org.simnet.networks
Class Elman

java.lang.Object
  org.simnet.interfaces.Network
      org.simnet.networks.Backprop
          org.simnet.networks.Elman

All Implemented Interfaces:

EventListener, WorldListener

public class Elman
extends Backprop

Elman networks are simple recurrent networks based on the work of Jeff Elman.

Field Summary

Fields inherited from class org.simnet.interfaces.Network

CONTINUOUS, DISCRETE

Constructor Summary

Elman()
Default constructor.

Method Summary

void	batchIterate() Batch train for one iteration.
protected void	buildInitialNetwork() Build the default network.
void	buildSnarliNetwork() Create the Snarli network.
void	defaultInit() Build network and initialize nodes and weights to appropriate values.
int	getNHidden()
int	getNInput()
void	iterate() Iterate Elman network learning.
void	randomize() Randomize Elman network (all but weights from hidden to copy).
void	train() Train Elman network.
void	update() Override update function to update input and copy layer first.
void	updateSimbrainNetwork() Update Simbrain network to match SNARLI network after it is trained.

Methods inherited from class org.simnet.networks.Backprop

attachInputsAndOutputs, batchTrain, getBiases, getDefaultNeuron, getEpochs, getError, getErrorInterval, getEta, getHid, getInp, getMu, getNInputs, getNOutputs, getOut, getTrainingINFile, getTrainingInputs, getTrainingOUTFile, getTrainingOutputs, setBiases, setEpochs, setError, setErrorInterval, setEta, setHid, setInp, setMu, setNHidden, setNInputs, setNOutputs, setOut, setTrainingINFile, setTrainingInputs, setTrainingOUTFile, setTrainingOutputs

Methods inherited from class org.simnet.interfaces.Network

addNetwork, addNetworkList, addNetworkListener, addNeuron, addNeuron, addNeuronList, addWeight, addWeight, changeNeuron, changeSynapse, checkAllBounds, clearActivations, clearInputs, close, deleteNetwork, deleteNeuron, deleteNeuron, deleteWeight, deleteWeight, fireClampChanged, fireCouplingChanged, fireNetworkChanged, fireNeuronAdded, fireNeuronChanged, fireNeuronDeleted, fireSubnetAdded, fireSubnetDeleted, fireSynapseAdded, fireSynapseChanged, fireSynapseDeleted, getClampNeurons, getClampWeights, getCouplingList, getDepth, getFlatNetworkList, getFlatNeuronList, getFlatSynapseList, getId, getIndents, getInputNeurons, getInteractionMode, getListenerList, getNetwork, getNetworkList, getNetworkParent, getNetworkThread, getNeuron, getNeuron, getNeuronCount, getNeuronList, getOutputNeurons, getPrecision, getRoot, getRoundingOff, getState, getSynapse, getTime, getTimeLabel, getTimeStep, getTimeType, getType, getUnits, getWeight, getWeight, getWeight, getWeightCount, getWeightList, getWorkspace, init, initNeurons, initParents, initWeights, isRoot, isRoundOffActivationValues, isUpdateCompleted, randomizeNeurons, randomizeWeights, removeNetworkListener, round, roundAll, setClampNeurons, setClampWeights, setId, setInteractionMode, setLowerBounds, setNetworkList, setNetworkParent, setNetworkThread, setNeuronList, setPrecision, setRoundingOff, setRoundOffActivationValues, setTime, setTimeStep, setUpdateCompleted, setUpperBounds, setWeightList, setWeightsToZero, setWorkspace, toString, updateAllNetworks, updateAllNeurons, updateAllWeights, updateIds, updateInputs, updateTime, updateTimeType, updateTopLevel, updateWorldListeners, updateWorlds, worldChanged

Methods inherited from class java.lang.Object

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

Constructor Detail

Elman

public Elman()

Default constructor.

Method Detail

defaultInit

public void defaultInit()

Build network and initialize nodes and weights to appropriate values.

Overrides:

defaultInit in class Backprop

buildInitialNetwork

protected void buildInitialNetwork()

Build the default network.

Overrides:

buildInitialNetwork in class Backprop

buildSnarliNetwork

public void buildSnarliNetwork()

Create the Snarli network. The Simbrain network is used . to set all weights and biases.

Overrides:

buildSnarliNetwork in class Backprop

train

public void train()

Train Elman network.

Overrides:

train in class Backprop

iterate

public void iterate()

Iterate Elman network learning.

Overrides:

iterate in class Backprop

batchIterate

public void batchIterate()

Batch train for one iteration.

Overrides:

batchIterate in class Backprop

updateSimbrainNetwork

public void updateSimbrainNetwork()

Update Simbrain network to match SNARLI network after it is trained.

Overrides:

updateSimbrainNetwork in class Backprop

randomize

public void randomize()

Randomize Elman network (all but weights from hidden to copy).

Overrides:

randomize in class Backprop

update

public void update()

Override update function to update input and copy layer first.

Overrides:

update in class Backprop

getNInput

public int getNInput()

Returns:

Number of input neurons.

getNHidden

public int getNHidden()

Overrides:

getNHidden in class Backprop

Returns:

Number of hidden neurons.