NearNeural

Class ReceptiveNeuron

java.lang.Object

NearNeural.ReceptiveNeuron

All Implemented Interfaces:

Neuron

public class ReceptiveNeuron
extends java.lang.Object
implements Neuron

A receptive Neuron is a form of Neuron that maintains Edges, which themselves point to other Neurons, be they static or themselves a form of Receptive Neuron. Receptive neurons can calculate their value through any means. Because the only common code is Edge management, this is all the Nodes do.

Version:

0.2.0a

Author:

Stephen Wattam

Field Summary

Fields

Modifier and Type	Field and Description
protected double	delta Stores the last value correction as per the backpropagation algorithm
protected Edge[]	edges Retains all Edges for this node.

Constructor Summary

Constructors

Constructor and Description
ReceptiveNeuron(Edge[] p_inputEdges, ThresholdingAlgorithm p_thresholdModel) Creates a Neuron with Edges from an array.
ReceptiveNeuron(ThresholdingAlgorithm p_thresholdModel) Creates a new neuron with no edges.

Method Summary

Methods

Modifier and Type	Method and Description
void	addEdge(Edge p_edge) Adds a single edge to the neuron.
void	addEdges(Neuron[] p_neuronList) Constructs randomly weighted edges to all neurons in the array provided.
void	calculateValue() pre-calculates the value of this neuron - this method requires that all previous layers have been run first in a feed-forward manner.
int	countEdges() Returns the number of edges currently in this neuron.
double	differentialValue() Returns the differential of the function that generates value at this point.
double	getDelta() Returns the last error value from backpropagation.
Edge	getEdge(int p_index) Returns an edge at the index given.
Edge[]	getInputEdges() Returns a list of all edges handled by this Neuron
boolean	isPresent(Edge p_edge) checks if an Edge is held in this layer.
void	prune(double p_threshold) Removes Edges with an absolute weight under the threshold given.
void	removeEdge(Edge p_edge) Removes an edge from this neuron.
void	removeEdge(int p_index) Removes an edge from the layer by index.
void	setDelta(double p_delta) Sets the difference between current value and desired value during the backpropagation process.
void	setEdgeWeight(int p_index, double p_weight) Sets the weight of an edge, as indexed by the connection to the current Neuron.
double	value() Returns the sum of all Edge values, multiplied by their weight.

Methods inherited from class java.lang.Object

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

Field Detail

edges

protected Edge[] edges

Retains all Edges for this node.

delta

protected double delta

Stores the last value correction as per the backpropagation algorithm

Constructor Detail

ReceptiveNeuron

public ReceptiveNeuron(ThresholdingAlgorithm p_thresholdModel)

Creates a new neuron with no edges.

Parameters:

p_thresholdModel - The thresholding algorithm to use

ReceptiveNeuron

public ReceptiveNeuron(Edge[] p_inputEdges,
               ThresholdingAlgorithm p_thresholdModel)

Creates a Neuron with Edges from an array. Generally speaking this will remain unused as all Neurons are supposed to link to all of the previous NeuronLayer's Neurons.

Parameters:

p_inputEdges - An array of edges to attach to the Neuron

p_thresholdModel - The thresholding algorithm model to use

Method Detail

addEdges

public void addEdges(Neuron[] p_neuronList)

Constructs randomly weighted edges to all neurons in the array provided.

Parameters:

p_neuronList - An array of neurons to create edges to

getDelta

public double getDelta()

Returns the last error value from backpropagation. This is used in the backpropagation process and can be used to indicate the success of the last training cycle, but is not stored across save/load events.

Returns:

The last target value adjustment which caused edges to be edited

setDelta

public void setDelta(double p_delta)

Sets the difference between current value and desired value during the backpropagation process. This can be edited manually but will be lost during save/load events and must be used in a specific order to make any difference to edge weights. This method does not edit edge weights directly, but can influence the backpropagation process of all layers prior to the current one.

Parameters:

p_delta - The delta value chosen for this neuron

setEdgeWeight

public void setEdgeWeight(int p_index,
                 double p_weight)
                   throws IndexOutOfBoundsException

Sets the weight of an edge, as indexed by the connection to the current Neuron.

Parameters:

p_index - The index of the Edge of which to edit the weight.

p_weight - The weight to set the Edge to.

Throws:

IndexOutOfBoundsException - in the event that the index given is not linked to an edge.

addEdge

public void addEdge(Edge p_edge)

Adds a single edge to the neuron.

Parameters:

p_edge - The edge to add.

removeEdge

public void removeEdge(Edge p_edge)
                throws ItemNotFoundException

Removes an edge from this neuron.

Parameters:

p_edge - The edge to remove.

Throws:

ItemNotFoundException

prune

public void prune(double p_threshold)

Removes Edges with an absolute weight under the threshold given. This is usually used after training for the final time and can be used to increase the speed of the net whilst sacrificing little in the way of accuracy.

Parameters:

p_threshold - The weighting under which to remove an Edge

removeEdge

public void removeEdge(int p_index)
                throws IndexOutOfBoundsException

Removes an edge from the layer by index.

Parameters:

p_index - The index of the edge to remove

Throws:

IndexOutOfBoundsException - in the event that the index given is out of range.

countEdges

public int countEdges()

Returns the number of edges currently in this neuron.

Returns:

The number of edges currently in the neuron

getInputEdges

public Edge[] getInputEdges()

Returns a list of all edges handled by this Neuron

value

public double value()

Returns the sum of all Edge values, multiplied by their weight.

Specified by:

value in interface Neuron

Returns:

The sum of all edges, which is the value for this Neuron.

differentialValue

public double differentialValue()

Returns the differential of the function that generates value at this point.

Returns:

The differential of the fucntion generating the valuye of this neuron

calculateValue

public void calculateValue()

pre-calculates the value of this neuron - this method requires that all previous layers have been run first in a feed-forward manner.

getEdge

public Edge getEdge(int p_index)
             throws IndexOutOfBoundsException

Returns an edge at the index given.

Parameters:

p_index - the index of the Edge to return

Returns:

The edge at the index given

Throws:

IndexOutOfBoundsException - in the event that the index given is invalid.

isPresent

public boolean isPresent(Edge p_edge)

checks if an Edge is held in this layer.

Parameters:

p_edge - The edge to check for

Returns:

true if the edge is present, false if it is not