org.simbrain.network.subnetworks

Class EchoStateNetwork

java.lang.Object

org.simbrain.network.groups.Group

org.simbrain.network.groups.Subnetwork

org.simbrain.network.subnetworks.EchoStateNetwork

public class EchoStateNetwork
extends Subnetwork

Builds an Echo-State Network with options for all valid weight configurations.

Author:

Zach Tosi

Constructor Summary

Constructors

Constructor and Description
EchoStateNetwork(Network network, int inputNodes, int reservoirNodes, int outputNodes, java.awt.geom.Point2D initialPosition) Constructor with size of layers specified.
EchoStateNetwork(Network network, java.awt.geom.Point2D initialPosition) Creates an empty ESN where neuron groups and synapse groups must be manually added.

Method Summary

Modifier and Type	Method and Description
void	connectLayers(Sparse inToRes, Sparse resRecurrent, Sparse outToRes) Connects all the layers of the network based on 3 connection objects each with their own connection parameters.
double[][]	getInputData()
NumericMatrix	getInputDataMatrix() Wrap input data in a DataMatrix Object.
NeuronGroup	getInputLayer()
Randomizer	getNoiseGenerator()
NeuronGroup	getOutputLayer()
NeuronGroup	getReservoirLayer()
double[][]	getTargetData()
NumericMatrix	getTargetDataMatrix() Wrap target data in a DataMatrix Object.
Network.TimeType	getTimeType()
Trainer	getTrainer() Return a trainer object that can be used to train this ESN.
java.lang.String	getUpdateMethodDesecription() Returns a description of this group's update method, which is displayed in the update manager panel.
boolean	getUseNoise()
void	initializeInputLayer(NeuronGroup neuronGroup) Initializes the input layer from a neuron group and adds it to the ESN.
void	initializeOutput(NeuronGroup neuronGroup) Initializes the output layer and adds it to the ESN.
void	initializeReservoir(NeuronGroup neuronGroup, SynapseGroup synapseGroup, double spectralRadius) Initializes the reservoir.
void	positionLayers() A helper method which positions the layers relative to each other in an aesthetically pleasing arrangement.
void	setBackWeights(boolean backWeights) Set to true for weights from output to reservoir.
void	setDirectInOutWeights(boolean directInOutWeights) Set to true for weights directly from input to output.
void	setInputData(double[][] inputData)
void	setNoise(boolean noise)
void	setNoiseGenerator(Randomizer noiseGenerator)
void	setRecurrentOutWeights(boolean recurrentWeights) Set to true for the output to receive input from itself from the previous time-step.
void	setReservoirNeuronType(NeuronUpdateRule reservoirNeuronType) Set type of reservoir neurons.
void	setSpectralRadius(double spectralRadius) Set spectral radius.
void	setTargetData(double[][] targetData)
void	setTimeType(Network.TimeType timeType)
void	setUseNoise(boolean noise)
void	update() Update this group.

Methods inherited from class org.simbrain.network.groups.Subnetwork

addAndLabelSynapseGroup, addNeuronGroup, addRowToTrainingSet, addSynapseGroup, clearActivations, connectNeuronGroups, connectNeuronGroups, connectNeuronGroups, delete, displayNeuronGroups, getEnabled, getFlatNeuronList, getFlatSynapseList, getIndexOfNeuronGroup, getLongDescription, getModifiableNeuronList, getNeuronGroup, getNeuronGroup, getNeuronGroupByLabel, getNeuronGroupCount, getNeuronGroupList, getNeuronGroupsAsList, getSynapseGroup, getSynapseGroup, getSynapseGroupByLabel, getSynapseGroupCount, getSynapseGroupList, isEmpty, recursivelySetIds, removeNeuronGroup, removeSynapseGroup, setDisplayNeuronGroups, setEnabled, size, toString

Methods inherited from class org.simbrain.network.groups.Group

getId, getLabel, getParentGroup, getParentNetwork, getStateInfo, hasParentGroup, isMarkedForDeletion, isTopLevelGroup, setId, setLabel, setMarkedForDeletion, setParentGroup, setStateInfo

Methods inherited from class java.lang.Object

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

Constructor Detail

EchoStateNetwork

public EchoStateNetwork(Network network,
                        int inputNodes,
                        int reservoirNodes,
                        int outputNodes,
                        java.awt.geom.Point2D initialPosition)

Constructor with size of layers specified.

Parameters:

network - the root network wherein ESNBuilder will build components

inputNodes - number of input nodes

reservoirNodes - number of reservoir nodes

outputNodes - number of output nodes

initialPosition - the initial position

EchoStateNetwork

public EchoStateNetwork(Network network,
                        java.awt.geom.Point2D initialPosition)

Creates an empty ESN where neuron groups and synapse groups must be manually added.

Parameters:

network -

initialPosition -

Method Detail

initializeInputLayer

public void initializeInputLayer(NeuronGroup neuronGroup)

Initializes the input layer from a neuron group and adds it to the ESN.

Parameters:

neuronGroup -

initializeReservoir

public void initializeReservoir(NeuronGroup neuronGroup,
                                SynapseGroup synapseGroup,
                                double spectralRadius)

Initializes the reservoir. This method must be given the recurrent synapse group in addition to the neuron group associated with the reservoir. This method scales the weights of the synapse group to conform the desired spectral radius and sets the spectral radius of this esn accordingly.

Parameters:

neuronGroup -

synapseGroup -

spectralRadius -

initializeOutput

public void initializeOutput(NeuronGroup neuronGroup)

Initializes the output layer and adds it to the ESN.

Parameters:

neuronGroup -

positionLayers

public void positionLayers()

A helper method which positions the layers relative to each other in an aesthetically pleasing arrangement.

update

public void update()

Update this group. Updates the ESN layer by layer, input then reservoir then output.

Overrides:

update in class Subnetwork

connectLayers

public void connectLayers(Sparse inToRes,
                          Sparse resRecurrent,
                          Sparse outToRes)

Connects all the layers of the network based on 3 connection objects each with their own connection parameters. Also scales the spectral radius of the recurrent connections in the reservoir.

Parameters:

inToRes - the connection object governing how the input connects to the reservoir.

resRecurrent - the connection object governing how the reservoir connects to itself.

outToRes - the connection object governing how the output connects to the reservoir. If these connections do not exist, pass null. If backWeights is false these connections will not be made regardless of whether or not outToRes is null.

getTrainer

public Trainer getTrainer()

Return a trainer object that can be used to train this ESN. The trainer has harvested state data as "inputs" and the desired outputs as outputs.

Returns:

the trainer.

setSpectralRadius

public void setSpectralRadius(double spectralRadius)

Set spectral radius.

Parameters:

spectralRadius - the spectral radius

setReservoirNeuronType

public void setReservoirNeuronType(NeuronUpdateRule reservoirNeuronType)

Set type of reservoir neurons.

Parameters:

reservoirNeuronType -

setBackWeights

public void setBackWeights(boolean backWeights)

Set to true for weights from output to reservoir.

Parameters:

backWeights -

setRecurrentOutWeights

public void setRecurrentOutWeights(boolean recurrentWeights)

Set to true for the output to receive input from itself from the previous time-step.

Parameters:

recurrentWeights -

setDirectInOutWeights

public void setDirectInOutWeights(boolean directInOutWeights)

Set to true for weights directly from input to output.

Parameters:

directInOutWeights - weights directly from input to output.

getInputLayer

public NeuronGroup getInputLayer()

Returns:

the inputLayer

getReservoirLayer

public NeuronGroup getReservoirLayer()

Returns:

the reservoirLayer

getOutputLayer

public NeuronGroup getOutputLayer()

Returns:

the outputLayer

getInputData

public double[][] getInputData()

Returns:

the inputData

setInputData

public void setInputData(double[][] inputData)

Parameters:

inputData - the inputData to set

getTargetData

public double[][] getTargetData()

Returns:

the targetData

setTargetData

public void setTargetData(double[][] targetData)

Parameters:

targetData - the targetData to set

getUseNoise

public boolean getUseNoise()

Returns:

the noise

setUseNoise

public void setUseNoise(boolean noise)

Parameters:

noise - the noise to set

getNoiseGenerator

public Randomizer getNoiseGenerator()

setNoiseGenerator

public void setNoiseGenerator(Randomizer noiseGenerator)

setNoise

public void setNoise(boolean noise)

getUpdateMethodDesecription

public java.lang.String getUpdateMethodDesecription()

Description copied from class: Group

Returns a description of this group's update method, which is displayed in the update manager panel.

Overrides:

getUpdateMethodDesecription in class Subnetwork

Returns:

a description of the update method.

getTimeType

public Network.TimeType getTimeType()

setTimeType

public void setTimeType(Network.TimeType timeType)

getInputDataMatrix

public NumericMatrix getInputDataMatrix()

Wrap input data in a DataMatrix Object.

Returns:

the data matrix containing this data.

getTargetDataMatrix

public NumericMatrix getTargetDataMatrix()

Wrap target data in a DataMatrix Object.

Returns:

the data matrix containing this data.