NearOCR

Class WeightedNode

java.lang.Object

NearOCR.WeightedNode

public class WeightedNode
extends java.lang.Object

Holds an adjusted weight for one output. A symbol contains many of these, one for each serialised net output node.

Version:

0.1

Author:

Stephen Wattam

Constructor Summary

Constructors

Constructor and Description
WeightedNode(int p_index, double p_target, double p_weight) Creates a new weighted node with a set index, weight and target.

Method Summary

Methods

Modifier and Type	Method and Description
void	adjustWeightByDeviation(double p_output) Adjusts the weight of this node such that its weight multiplied by the given output is exactly equal to the target.
double	getDeviation(double p_output) Returns the difference between the weight-adjusted output neuron value and the internally stored target.
int	getIndex() Returns the current index of this node.
double	getTarget() Returns the current target value, between 0 and 1, of this weighted node.
double	getWeight() Returns the weight this node currently holds.
void	setIndex(int p_index) Sets the index number of this node.
void	setTarget(double p_target) Sets the target value, from which error is calculated
void	setWeight(double p_weight) Sets the weight of this node, by which the output of the net will be multiplied in order to fin its true output.
java.lang.String	toString() Returns this node represented as a string.

Methods inherited from class java.lang.Object

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

Constructor Detail

WeightedNode

public WeightedNode(int p_index,
            double p_target,
            double p_weight)

Creates a new weighted node with a set index, weight and target.

Parameters:

p_index - The index number of the node

p_target - The target value of the node, initially 0

p_weight - The weight value of the node, initially 1 to preserver net output

Method Detail

setTarget

public void setTarget(double p_target)

Sets the target value, from which error is calculated

Parameters:

p_target - The value to set as a target, from 0 to 1 inclusive

Throws:

ValueOutOfBoundsException - Thrown in the event that the value given in outside of the permissable range

getTarget

public double getTarget()

Returns the current target value, between 0 and 1, of this weighted node.

Returns:

The target value

setIndex

public void setIndex(int p_index)

Sets the index number of this node.

Parameters:

p_index - the index to set this node to.

getIndex

public int getIndex()

Returns the current index of this node.

Returns:

The current index of this node.

setWeight

public void setWeight(double p_weight)

Sets the weight of this node, by which the output of the net will be multiplied in order to fin its true output.

Parameters:

p_weight - The weight to set

getWeight

public double getWeight()

Returns the weight this node currently holds.

Returns:

The current weight of the node

getDeviation

public double getDeviation(double p_output)

Returns the difference between the weight-adjusted output neuron value and the internally stored target. This takes the form deviation == target-(weight*p_output)

Parameters:

p_output - The output neuron's value

Returns:

The deviation for this node only

adjustWeightByDeviation

public void adjustWeightByDeviation(double p_output)

Adjusts the weight of this node such that its weight multiplied by the given output is exactly equal to the target. This is used to compensate for the unique fingerprints of each net, removing inaccuracy whilst still allowing for a weighting system to judge confidence values.

Parameters:

p_output - The output from the net stack which matches this node's index number.

toString

public java.lang.String toString()

Returns this node represented as a string. This is used in the save/load process.

Overrides:

toString in class java.lang.Object

Returns:

A string representing this object.