org.encog.neural.neat.training.opp

Class NEATMutation

java.lang.Object

org.encog.neural.neat.training.opp.NEATMutation

All Implemented Interfaces:

EvolutionaryOperator

Direct Known Subclasses:

NEATMutateAddLink, NEATMutateAddNode, NEATMutateRemoveLink, NEATMutateWeights

public abstract class NEATMutation
extends Object
implements EvolutionaryOperator

This class represents a NEAT mutation. NEAT supports several different types of mutations. This class provides common utility needed by any sort of a NEAT mutation. This class is abstract and cannot be used by itself. ----------------------------------------------------------------------------- http://www.cs.ucf.edu/~kstanley/ Encog's NEAT implementation was drawn from the following three Journal Articles. For more complete BibTeX sources, see NEATNetwork.java. Evolving Neural Networks Through Augmenting Topologies Generating Large-Scale Neural Networks Through Discovering Geometric Regularities Automatic feature selection in neuroevolution

Constructor Summary

Constructors

Constructor and Description
NEATMutation()

Method Summary

Methods

Modifier and Type	Method and Description
NEATNeuronGene	chooseRandomNeuron(NEATGenome target, boolean choosingFrom) Choose a random neuron.
void	createLink(NEATGenome target, long neuron1ID, long neuron2ID, double weight) Create a link between two neuron id's.
int	getElementPos(NEATGenome target, long neuronID) Get the specified neuron's index.
EvolutionaryAlgorithm	getOwner()
void	init(EvolutionaryAlgorithm theOwner) Called to setup the evolutionary operator.
boolean	isDuplicateLink(NEATGenome target, long fromNeuronID, long toNeuronID) Determine if this is a duplicate link.
boolean	isNeuronNeeded(NEATGenome target, long neuronID) Determines if a neuron is still needed.
NEATGenome	obtainGenome(Genome[] parents, int parentIndex, Genome[] offspring, int offspringIndex) Obtain the NEATGenome that we will mutate.
int	offspringProduced()
int	parentsNeeded()
void	removeNeuron(NEATGenome target, long neuronID) Remove the specified neuron.

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.encog.ml.ea.opp.EvolutionaryOperator

performOperation

Constructor Detail

NEATMutation

public NEATMutation()

Method Detail

chooseRandomNeuron

public NEATNeuronGene chooseRandomNeuron(NEATGenome target,
                                boolean choosingFrom)

Choose a random neuron.

Parameters:

target - The target genome. Should the input and bias neurons be included.

choosingFrom - True if we are chosing from all neurons, false if we exclude the input and bias.

Returns:

The random neuron.

createLink

public void createLink(NEATGenome target,
              long neuron1ID,
              long neuron2ID,
              double weight)

Create a link between two neuron id's. Create or find any necessary innovation records.

Parameters:

target - The target genome.

neuron1ID - The id of the source neuron.

neuron2ID - The id of the target neuron.

weight - The weight of this new link.

getElementPos

public int getElementPos(NEATGenome target,
                long neuronID)

Get the specified neuron's index.

Parameters:

neuronID - The neuron id to check for.

Returns:

The index.

getOwner

public EvolutionaryAlgorithm getOwner()

Returns:

the owner

init

public void init(EvolutionaryAlgorithm theOwner)

Called to setup the evolutionary operator.

Specified by:

init in interface EvolutionaryOperator

Parameters:

theOwner - The evolutionary algorithm used with this operator.

isDuplicateLink

public boolean isDuplicateLink(NEATGenome target,
                      long fromNeuronID,
                      long toNeuronID)

Determine if this is a duplicate link.

Parameters:

fromNeuronID - The from neuron id.

toNeuronID - The to neuron id.

Returns:

True if this is a duplicate link.

isNeuronNeeded

public boolean isNeuronNeeded(NEATGenome target,
                     long neuronID)

Determines if a neuron is still needed. If all links to/from a neuron have been removed, then the neuron is no longer needed.

Parameters:

target - The target genome.

neuronID - The neuron id to check for.

Returns:

Returns true, if the neuron is still needed.

obtainGenome

public NEATGenome obtainGenome(Genome[] parents,
                      int parentIndex,
                      Genome[] offspring,
                      int offspringIndex)

Obtain the NEATGenome that we will mutate. NEAT mutates the genomes in place. So the parent and child genome must be the same literal object. Throw an exception, if this is not the case.

Parameters:

parents - The parents.

parentIndex - The parent index.

offspring - The offspring.

offspringIndex - The offspring index.

Returns:

The genome that we will mutate.

offspringProduced

public int offspringProduced()

Specified by:

offspringProduced in interface EvolutionaryOperator

Returns:

Returns 1, as NEAT mutations only produce one child.

parentsNeeded

public int parentsNeeded()

Specified by:

parentsNeeded in interface EvolutionaryOperator

Returns:

Returns 1, as mutations typically are asexual and only require a single parent.

removeNeuron

public void removeNeuron(NEATGenome target,
                long neuronID)

Remove the specified neuron.

Parameters:

target - The target genome.

neuronID - The neuron to remove.