NearNeural

Class HyperbolicTangentThresholdModel

java.lang.Object

NearNeural.WeightedThresholdModel

NearNeural.HyperbolicTangentThresholdModel

All Implemented Interfaces:

ThresholdingAlgorithm

public class HyperbolicTangentThresholdModel
extends WeightedThresholdModel
implements ThresholdingAlgorithm

Creates a new thresholding algorithm based on the hyperbolic tangent function. This approximates sigmoid closely only with a slightly smoother curve at k~1, and hence can offer slightly better training results.

Version:

0.3

Author:

Stephen Wattam (differentiatioon by Craig Edwards)

Field Summary

Fields inherited from class NearNeural.WeightedThresholdModel

k

Constructor Summary

Constructors

Constructor and Description
HyperbolicTangentThresholdModel(double p_k) Creates a new tangental function with a given steepness.

Method Summary

Methods

Modifier and Type	Method and Description
double	differentialValue(Edge[] edges) The value as run through the differential of the current function.
boolean	validateInput(double input) Validates input against the applicable range for this thresholding algorithm.
double	value(Edge[] edges) Calculates Neuron value from all Edges and their source Neurons.

Methods inherited from class NearNeural.WeightedThresholdModel

getK, setK

Methods inherited from class java.lang.Object

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

Constructor Detail

HyperbolicTangentThresholdModel

public HyperbolicTangentThresholdModel(double p_k)

Creates a new tangental function with a given steepness. K is used as a coefficient of x in the hyperbolic tangent function.

Parameters:

p_k - The steepness value

Method Detail

value

public double value(Edge[] edges)

Calculates Neuron value from all Edges and their source Neurons.

Specified by:

value in interface ThresholdingAlgorithm

Parameters:

edges - All of this neuron's edges

Returns:

This Neuron's value

See Also:

ThresholdingAlgorithm.value(Edge[])

validateInput

public boolean validateInput(double input)

Validates input against the applicable range for this thresholding algorithm.

Specified by:

validateInput in interface ThresholdingAlgorithm

Parameters:

input - The value to check

Returns:

True if the value is applicable to run though without error

See Also:

ThresholdingAlgorithm.validateInput(double)

differentialValue

public double differentialValue(Edge[] edges)

The value as run through the differential of the current function. This may not be used during backpropagation but is used in most forms of gradient-descent backpropagation.

Specified by:

differentialValue in interface ThresholdingAlgorithm

Parameters:

edges - All edges going into this neurons

Returns:

The differential value