/*
* Encog(tm) Core v3.4 - Java Version
* http://www.heatonresearch.com/encog/
* https://github.com/encog/encog-java-core
* Copyright 2008-2016 Heaton Research, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
* For more information on Heaton Research copyrights, licenses
* and trademarks visit:
* http://www.heatonresearch.com/copyright
*/
package org.encog.neural.neat;
import java.io.Serializable;
import java.util.Random;
import org.encog.Encog;
import org.encog.engine.network.activation.ActivationFunction;
import org.encog.engine.network.activation.ActivationSteepenedSigmoid;
import org.encog.mathutil.randomize.factory.RandomFactory;
import org.encog.ml.MLError;
import org.encog.ml.MLRegression;
import org.encog.ml.data.MLData;
import org.encog.ml.data.MLDataSet;
import org.encog.ml.ea.codec.GeneticCODEC;
import org.encog.ml.ea.genome.Genome;
import org.encog.ml.ea.population.BasicPopulation;
import org.encog.ml.ea.species.BasicSpecies;
import org.encog.neural.NeuralNetworkError;
import org.encog.neural.hyperneat.FactorHyperNEATGenome;
import org.encog.neural.hyperneat.HyperNEATCODEC;
import org.encog.neural.hyperneat.HyperNEATGenome;
import org.encog.neural.hyperneat.substrate.Substrate;
import org.encog.neural.neat.training.NEATGenome;
import org.encog.neural.neat.training.NEATInnovationList;
import org.encog.util.identity.BasicGenerateID;
import org.encog.util.identity.GenerateID;
import org.encog.util.obj.ChooseObject;
/**
* A population for a NEAT or HyperNEAT system. This population holds the
* genomes, substrate and other values for a NEAT or HyperNEAT network.
* -----------------------------------------------------------------------------
* 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
*/
public class NEATPopulation extends BasicPopulation implements Serializable,
MLError, MLRegression {
/**
* The default survival rate.
*/
public static final double DEFAULT_SURVIVAL_RATE = 0.2;
/**
* The activation function to use.
*/
public static final String PROPERTY_NEAT_ACTIVATION = "neatAct";
/**
* Property tag for the population size.
*/
public static final String PROPERTY_POPULATION_SIZE = "populationSize";
/**
* Property tag for the survival rate.
*/
public static final String PROPERTY_SURVIVAL_RATE = "survivalRate";
/**
* Serial id.
*/
private static final long serialVersionUID = 1L;
/**
* Default number of activation cycles.
*/
public static final int DEFAULT_CYCLES = 4;
/**
* Property to hold the number of cycles.
*/
public static final String PROPERTY_CYCLES = "cycles";
/**
* Default link weight range for NEAT networks.
*/
public static final double DEFAULT_NEAT_WEIGHT_RANGE = 1.0;
/**
* Default link weight range for HyperNEAT networks.
*/
public static final double DEFAULT_HYPERNEAT_WEIGHT_RANGE = 5.0;
/**
* Change the weight, do not allow the weight to go out of the weight range.
*
* @param w
* The amount to change the weight by.
* @param weightRange
* Specify the weight range. The range is from -weightRange to
* +weightRange.
* @return The new weight value.
*/
public static double clampWeight(final double w, final double weightRange) {
if (w < -weightRange) {
return -weightRange;
} else if (w > weightRange) {
return weightRange;
} else {
return w;
}
}
/**
* The number of activation cycles that the networks produced by this
* population will use.
*/
private int activationCycles = NEATPopulation.DEFAULT_CYCLES;
/**
* Generate gene id's.
*/
private final GenerateID geneIDGenerate = new BasicGenerateID();
/**
* Generate innovation id's.
*/
private final GenerateID innovationIDGenerate = new BasicGenerateID();
/**
* A list of innovations, or null if this feature is not being used.
*/
private NEATInnovationList innovations;
/**
* The weight range. Weights will be between -weight and +weight.
*/
private double weightRange = DEFAULT_NEAT_WEIGHT_RANGE;
/**
* The best genome that we've currently decoded into the bestNetwork
* property. If this value changes to point to a new genome reference then
* the phenome will need to be recalculated.
*/
private Genome cachedBestGenome;
/**
* The best network. If the population is used as an MLMethod, then this
* network will represent.
*/
private NEATNetwork bestNetwork;
/**
* The number of input units. All members of the population must agree with
* this number.
*/
int inputCount;
/**
* The number of output units. All members of the population must agree with
* this number.
*/
int outputCount;
/**
* The survival rate.
*/
private double survivalRate = NEATPopulation.DEFAULT_SURVIVAL_RATE;
/**
* The substrate, if this is a hyperneat network.
*/
private Substrate substrate;
/**
* The activation functions that we can choose from.
*/
private final ChooseObject<ActivationFunction> activationFunctions = new ChooseObject<ActivationFunction>();
/**
* The CODEC used to decode the NEAT genomes into networks. Different
* CODEC's are used for NEAT vs HyperNEAT.
*/
private GeneticCODEC codec;
/**
* The initial connection density for the initial random population of
* genomes.
*/
private double initialConnectionDensity = 0.1;
/**
* A factory to create random number generators.
*/
private RandomFactory randomNumberFactory = Encog.getInstance()
.getRandomFactory().factorFactory();
/**
* An empty constructor for serialization.
*/
public NEATPopulation() {
}
/**
* Construct a starting NEAT population. This does not generate the initial
* random population of genomes.
*
* @param inputCount
* The input neuron count.
* @param outputCount
* The output neuron count.
* @param populationSize
* The population size.
*/
public NEATPopulation(final int inputCount, final int outputCount,
final int populationSize) {
super(populationSize, null);
this.inputCount = inputCount;
this.outputCount = outputCount;
setNEATActivationFunction(new ActivationSteepenedSigmoid());
if (populationSize == 0) {
throw new NeuralNetworkError(
"Population must have more than zero genomes.");
}
}
/**
* Construct a starting HyperNEAT population. This does not generate the
* initial random population of genomes.
*
* @param theSubstrate
* The substrate ID.
* @param populationSize The population size.
*/
public NEATPopulation(final Substrate theSubstrate, final int populationSize) {
super(populationSize, new FactorHyperNEATGenome());
this.substrate = theSubstrate;
this.inputCount = 6;
this.outputCount = 2;
this.weightRange = DEFAULT_HYPERNEAT_WEIGHT_RANGE;
HyperNEATGenome.buildCPPNActivationFunctions(this.activationFunctions);
}
/**
* @return A newly generated gene id.
*/
public long assignGeneID() {
return this.geneIDGenerate.generate();
}
/**
* @return A newly generated innovation id.
*/
public long assignInnovationID() {
return this.innovationIDGenerate.generate();
}
/**
* {@inheritDoc}
*/
@Override
public double calculateError(final MLDataSet data) {
updateBestNetwork();
if( this.bestNetwork==null ) {
return Double.POSITIVE_INFINITY;
}
return this.bestNetwork.calculateError(data);
}
/**
* {@inheritDoc}
*/
@Override
public MLData compute(final MLData input) {
updateBestNetwork();
return this.bestNetwork.compute(input);
}
/**
* @return Get the activation cycles.
*/
public int getActivationCycles() {
return this.activationCycles;
}
/**
* @return the activationFunctions
*/
public ChooseObject<ActivationFunction> getActivationFunctions() {
return this.activationFunctions;
}
/**
* @return the codec
*/
public GeneticCODEC getCODEC() {
return this.codec;
}
/**
* @return the geneIDGenerate
*/
public GenerateID getGeneIDGenerate() {
return this.geneIDGenerate;
}
/**
* {@inheritDoc}
*/
@Override
public NEATGenomeFactory getGenomeFactory() {
return (NEATGenomeFactory) super.getGenomeFactory();
}
/**
* @return the initialConnectionDensity
*/
public double getInitialConnectionDensity() {
return this.initialConnectionDensity;
}
/**
* @return the innovationIDGenerate
*/
public GenerateID getInnovationIDGenerate() {
return this.innovationIDGenerate;
}
/**
* @return Get the innovations.
*/
public NEATInnovationList getInnovations() {
return this.innovations;
}
/**
* {@inheritDoc}
*/
@Override
public int getInputCount() {
return this.inputCount;
}
/**
* {@inheritDoc}
*/
@Override
public int getOutputCount() {
return this.outputCount;
}
/**
* @return the randomNumberFactory
*/
public RandomFactory getRandomNumberFactory() {
return this.randomNumberFactory;
}
/**
* @return Returns the hyper-neat substrate.
*/
public Substrate getSubstrate() {
return this.substrate;
}
/**
* @return The survival rate, this is the number of genomes used to mate.
*/
public double getSurvivalRate() {
return this.survivalRate;
}
/**
* @return the weightRange
*/
public double getWeightRange() {
return this.weightRange;
}
/**
* @return Returns true if this is a hyperneat population.
*/
public boolean isHyperNEAT() {
return this.substrate != null;
}
/**
* Create an initial random population.
*/
public void reset() {
// create the genome factory
if (isHyperNEAT()) {
this.codec = new HyperNEATCODEC();
setGenomeFactory(new FactorHyperNEATGenome());
} else {
this.codec = new NEATCODEC();
setGenomeFactory(new FactorNEATGenome());
}
// create the new genomes
getSpecies().clear();
// reset counters
getGeneIDGenerate().setCurrentID(1);
getInnovationIDGenerate().setCurrentID(1);
final Random rnd = this.randomNumberFactory.factor();
// create one default species
final BasicSpecies defaultSpecies = new BasicSpecies();
defaultSpecies.setPopulation(this);
// create the initial population
for (int i = 0; i < getPopulationSize(); i++) {
final NEATGenome genome = getGenomeFactory().factor(rnd, this,
this.inputCount, this.outputCount,
this.initialConnectionDensity);
defaultSpecies.add(genome);
}
defaultSpecies.setLeader(defaultSpecies.getMembers().get(0));
getSpecies().add(defaultSpecies);
// create initial innovations
setInnovations(new NEATInnovationList(this));
}
/**
* Set the number of activation cycles to use.
* @param activationCycles The number of activatino cycles to use.
*/
public void setActivationCycles(final int activationCycles) {
this.activationCycles = activationCycles;
}
/**
* @param codec
* the codec to set
*/
public void setCODEC(final GeneticCODEC codec) {
this.codec = codec;
}
/**
* @param initialConnectionDensity
* the initialConnectionDensity to set
*/
public void setInitialConnectionDensity(
final double initialConnectionDensity) {
this.initialConnectionDensity = initialConnectionDensity;
}
/**
* Set the innovation list to use.
* @param theInnovations The innovation list to use.
*/
public void setInnovations(final NEATInnovationList theInnovations) {
this.innovations = theInnovations;
}
/**
* @param inputCount
* the inputCount to set
*/
public void setInputCount(final int inputCount) {
this.inputCount = inputCount;
}
/**
* Specify to use a single activation function. This is typically the case
* for NEAT, but not for HyperNEAT.
*
* @param af The activation function to use.
*/
public void setNEATActivationFunction(final ActivationFunction af) {
this.activationFunctions.clear();
this.activationFunctions.add(1.0, af);
this.activationFunctions.finalizeStructure();
}
/**
* @param outputCount
* the outputCount to set
*/
public void setOutputCount(final int outputCount) {
this.outputCount = outputCount;
}
/**
* @param randomNumberFactory
* the randomNumberFactory to set
*/
public void setRandomNumberFactory(final RandomFactory randomNumberFactory) {
this.randomNumberFactory = randomNumberFactory;
}
/**
* @param substrate
* the substrate to set
*/
public void setSubstrate(final Substrate substrate) {
this.substrate = substrate;
}
/**
* Set the survival rate, this is the percent of the population allowed to mate.
* @param theSurvivalRate The survival rate.
*/
public void setSurvivalRate(final double theSurvivalRate) {
this.survivalRate = theSurvivalRate;
}
/**
* Sets the weight range for links in genomes in this population.
* @param weightRange The link weight range.
*/
public void setWeightRange(final double weightRange) {
this.weightRange = weightRange;
}
/**
* See if the best genome has changed, and decode a new best network, if
* needed.
*/
private void updateBestNetwork() {
if (getBestGenome() != this.cachedBestGenome) {
this.cachedBestGenome = getBestGenome();
this.bestNetwork = (NEATNetwork) getCODEC().decode(getBestGenome());
}
}
}