org.apache.commons.math3.ml.neuralnet.sofm

Class LearningFactorFunctionFactory

java.lang.Object

org.apache.commons.math3.ml.neuralnet.sofm.LearningFactorFunctionFactory

public class LearningFactorFunctionFactory
extends java.lang.Object

Factory for creating instances of LearningFactorFunction.

Since:

3.3

Constructor Summary

Constructors

Modifier	Constructor and Description
private	LearningFactorFunctionFactory() Class contains only static methods.

Method Summary

Methods

Modifier and Type	Method and Description
static LearningFactorFunction	exponentialDecay(double initValue, double valueAtNumCall, long numCall) Creates an exponential decay function.
static LearningFactorFunction	quasiSigmoidDecay(double initValue, double slope, long numCall) Creates an sigmoid-like LearningFactorFunction function.

Methods inherited from class java.lang.Object

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

Constructor Detail

LearningFactorFunctionFactory

private LearningFactorFunctionFactory()

Class contains only static methods.

Method Detail

exponentialDecay

public static LearningFactorFunction exponentialDecay(double initValue,
                                      double valueAtNumCall,
                                      long numCall)

Creates an exponential decay function. It will compute a e-x / b, where x is the (integer) independent variable and

• a = initValue
• b = -numCall / ln(valueAtNumCall / initValue)

Parameters:

initValue - Initial value, i.e. value(0).

valueAtNumCall - Value of the function at numCall.

numCall - Argument for which the function returns valueAtNumCall.

Returns:

the learning factor function.

Throws:

OutOfRangeException - if initValue <= 0 or initValue > 1.

NotStrictlyPositiveException - if valueAtNumCall <= 0.

NumberIsTooLargeException - if valueAtNumCall >= initValue.

NotStrictlyPositiveException - if numCall <= 0.

quasiSigmoidDecay

public static LearningFactorFunction quasiSigmoidDecay(double initValue,
                                       double slope,
                                       long numCall)

Creates an sigmoid-like LearningFactorFunction function. The function f will have the following properties:

• f(0) = initValue
• numCall is the inflexion point
• slope = f'(numCall)

Parameters:

initValue - Initial value, i.e. value(0).

slope - Value of the function derivative at numCall.

numCall - Inflexion point.

Returns:

the learning factor function.

Throws:

OutOfRangeException - if initValue <= 0 or initValue > 1.

NumberIsTooLargeException - if slope >= 0.

NotStrictlyPositiveException - if numCall <= 0.