com.kylecorry.lann

Class GeneticTrainingAlgorithm

java.lang.Object

com.kylecorry.lann.GeneticTrainingAlgorithm

public class GeneticTrainingAlgorithm
extends java.lang.Object

Constructor Summary

Constructors

Constructor and Description
GeneticTrainingAlgorithm(int[] topology, int numGenes, double learningRate, double mutation, boolean classify, boolean keepBest) Train the neural networks using a version of a genetic training algorithm.

Method Summary

Modifier and Type	Method and Description
NeuralNetwork	best(boolean max) Get the best neural network based on its fitness.
double	bestFitness(boolean max) Get the best fitness in the generation.
void	evolve(boolean max) Evolve the neural networks to create the next generation.
double	getFitness(NeuralNetwork net) Get the fitness of a neural network.
java.util.HashMap<NeuralNetwork,java.lang.Double>	getGenes() Get a HashMap linking the neural networks to their fitness.
void	load(java.lang.String filename) Load a gene pool from a file.
void	setFitness(NeuralNetwork net, double score) Set the fitness of a neural network.

Methods inherited from class java.lang.Object

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

Constructor Detail

GeneticTrainingAlgorithm

public GeneticTrainingAlgorithm(int[] topology,
                                int numGenes,
                                double learningRate,
                                double mutation,
                                boolean classify,
                                boolean keepBest)

Train the neural networks using a version of a genetic training algorithm.

Parameters:

topology - The topology of the networks in this format int[]{input, hidden1, hidden2, hiddenN, output}

numGenes - The number of different genes per generation.

learningRate - The amount a mutation affects the genes of the next generation.

mutation - The percent chance of a mutation occurring: a double from 0..1 inclusive

classify - A boolean flag specifying whether this is a classification problem.

keepBest - A boolean flag specifying whether to keep the gene with the best fitness in the next generation without mutation.

Method Detail

getGenes

public java.util.HashMap<NeuralNetwork,java.lang.Double> getGenes()

Get a HashMap linking the neural networks to their fitness.

Returns:

A HashMap linking the nets with their fitness.

load

public void load(java.lang.String filename)

Load a gene pool from a file.

Parameters:

filename - The filename of the file containing the weight values for a neural net.

setFitness

public void setFitness(NeuralNetwork net,
                       double score)

Set the fitness of a neural network.

Parameters:

net - The neural network being evaluated.

score - The fitness of the neural network.

getFitness

public double getFitness(NeuralNetwork net)

Get the fitness of a neural network.

Parameters:

net - The neural network being evaluated.

Returns:

The fitness of the neural network.

evolve

public void evolve(boolean max)

Evolve the neural networks to create the next generation.

Parameters:

max - A boolean flag of whether to maximize or minimize the fitness. Max is true, min is false.

best

public NeuralNetwork best(boolean max)

Get the best neural network based on its fitness.

Parameters:

max - A boolean flag of whether a higher fitness is better than a lower fitness. true for max, false for min.

Returns:

The best neural network based on its fitness.

bestFitness

public double bestFitness(boolean max)

Get the best fitness in the generation.

Parameters:

max - A boolean flag of whether a higher fitness is better than a lower fitness. true for max, false for min.

Returns:

The fitness of the best neural network.