/*
* 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.InputStream;
import java.io.OutputStream;
import java.util.List;
import java.util.Map;
import org.encog.engine.network.activation.ActivationFunction;
import org.encog.ml.ea.genome.Genome;
import org.encog.ml.ea.species.BasicSpecies;
import org.encog.ml.ea.species.Species;
import org.encog.neural.hyperneat.FactorHyperNEATGenome;
import org.encog.neural.hyperneat.HyperNEATCODEC;
import org.encog.neural.hyperneat.HyperNEATGenome;
import org.encog.neural.neat.training.NEATGenome;
import org.encog.neural.neat.training.NEATInnovation;
import org.encog.neural.neat.training.NEATInnovationList;
import org.encog.neural.neat.training.NEATLinkGene;
import org.encog.neural.neat.training.NEATNeuronGene;
import org.encog.persist.EncogFileSection;
import org.encog.persist.EncogPersistor;
import org.encog.persist.EncogReadHelper;
import org.encog.persist.EncogWriteHelper;
import org.encog.persist.PersistConst;
import org.encog.util.csv.CSVFormat;
/**
* Persist 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 PersistNEATPopulation implements EncogPersistor {
/**
* Type for the Compositional pattern-producing networks used by HyperNEAT.
*/
public static final String TYPE_CPPN = "cppn";
/**
* Convert a NEATNeuronType enum to a string.
* @param t The type.
* @return The string type.
*/
public static String neuronTypeToString(final NEATNeuronType t) {
switch (t) {
case Bias:
return "b";
case Hidden:
return "h";
case Input:
return "i";
case None:
return "n";
case Output:
return "o";
default:
return null;
}
}
public static NEATNeuronType stringToNeuronType(final String t) {
if (t.equals("b")) {
return NEATNeuronType.Bias;
} else if (t.equals("h")) {
return NEATNeuronType.Hidden;
} else if (t.equals("i")) {
return NEATNeuronType.Input;
} else if (t.equals("n")) {
return NEATNeuronType.None;
} else if (t.equals("o")) {
return NEATNeuronType.Output;
} else {
return null;
}
}
@Override
public int getFileVersion() {
return 1;
}
@Override
public String getPersistClassString() {
return NEATPopulation.class.getSimpleName();
}
@Override
public Object read(final InputStream is) {
long nextInnovationID = 0;
long nextGeneID = 0;
final NEATPopulation result = new NEATPopulation();
final NEATInnovationList innovationList = new NEATInnovationList();
innovationList.setPopulation(result);
result.setInnovations(innovationList);
final EncogReadHelper in = new EncogReadHelper(is);
EncogFileSection section;
while ((section = in.readNextSection()) != null) {
if (section.getSectionName().equals("NEAT-POPULATION")
&& section.getSubSectionName().equals("INNOVATIONS")) {
for (final String line : section.getLines()) {
final List<String> cols = EncogFileSection
.splitColumns(line);
final NEATInnovation innovation = new NEATInnovation();
final int innovationID = Integer.parseInt(cols.get(1));
innovation.setInnovationID(innovationID);
innovation.setNeuronID(Integer.parseInt(cols.get(2)));
result.getInnovations().getInnovations()
.put(cols.get(0), innovation);
nextInnovationID = Math.max(nextInnovationID,
innovationID + 1);
}
} else if (section.getSectionName().equals("NEAT-POPULATION")
&& section.getSubSectionName().equals("SPECIES")) {
NEATGenome lastGenome = null;
BasicSpecies lastSpecies = null;
for (final String line : section.getLines()) {
final List<String> cols = EncogFileSection
.splitColumns(line);
if (cols.get(0).equalsIgnoreCase("s")) {
lastSpecies = new BasicSpecies();
lastSpecies.setPopulation(result);
lastSpecies.setAge(Integer.parseInt(cols.get(1)));
lastSpecies.setBestScore(CSVFormat.EG_FORMAT.parse(cols
.get(2)));
lastSpecies.setGensNoImprovement(Integer.parseInt(cols
.get(3)));
result.getSpecies().add(lastSpecies);
} else if (cols.get(0).equalsIgnoreCase("g")) {
final boolean isLeader = lastGenome == null;
lastGenome = new NEATGenome();
lastGenome.setInputCount(result.getInputCount());
lastGenome.setOutputCount(result.getOutputCount());
lastGenome.setSpecies(lastSpecies);
lastGenome.setAdjustedScore(CSVFormat.EG_FORMAT
.parse(cols.get(1)));
lastGenome.setScore(CSVFormat.EG_FORMAT.parse(cols
.get(2)));
lastGenome.setBirthGeneration(Integer.parseInt(cols
.get(3)));
lastSpecies.add(lastGenome);
if (isLeader) {
lastSpecies.setLeader(lastGenome);
}
} else if (cols.get(0).equalsIgnoreCase("n")) {
final NEATNeuronGene neuronGene = new NEATNeuronGene();
final int geneID = Integer.parseInt(cols.get(1));
neuronGene.setId(geneID);
final ActivationFunction af = EncogFileSection
.parseActivationFunction(cols.get(2));
neuronGene.setActivationFunction(af);
neuronGene.setNeuronType(PersistNEATPopulation
.stringToNeuronType(cols.get(3)));
neuronGene
.setInnovationId(Integer.parseInt(cols.get(4)));
lastGenome.getNeuronsChromosome().add(neuronGene);
nextGeneID = Math.max(geneID + 1, nextGeneID);
} else if (cols.get(0).equalsIgnoreCase("l")) {
final NEATLinkGene linkGene = new NEATLinkGene();
linkGene.setId(Integer.parseInt(cols.get(1)));
linkGene.setEnabled(Integer.parseInt(cols.get(2)) > 0);
linkGene.setFromNeuronID(Integer.parseInt(cols.get(3)));
linkGene.setToNeuronID(Integer.parseInt(cols.get(4)));
linkGene.setWeight(CSVFormat.EG_FORMAT.parse(cols
.get(5)));
linkGene.setInnovationId(Integer.parseInt(cols.get(6)));
lastGenome.getLinksChromosome().add(linkGene);
}
}
} else if (section.getSectionName().equals("NEAT-POPULATION")
&& section.getSubSectionName().equals("CONFIG")) {
final Map<String, String> params = section.parseParams();
final String afStr = params
.get(NEATPopulation.PROPERTY_NEAT_ACTIVATION);
if (afStr.equalsIgnoreCase(PersistNEATPopulation.TYPE_CPPN)) {
HyperNEATGenome.buildCPPNActivationFunctions(result
.getActivationFunctions());
} else {
result.setNEATActivationFunction(EncogFileSection
.parseActivationFunction(params,
NEATPopulation.PROPERTY_NEAT_ACTIVATION));
}
result.setActivationCycles(EncogFileSection.parseInt(params,
PersistConst.ACTIVATION_CYCLES));
result.setInputCount(EncogFileSection.parseInt(params,
PersistConst.INPUT_COUNT));
result.setOutputCount(EncogFileSection.parseInt(params,
PersistConst.OUTPUT_COUNT));
result.setPopulationSize(EncogFileSection.parseInt(params,
NEATPopulation.PROPERTY_POPULATION_SIZE));
result.setSurvivalRate(EncogFileSection.parseDouble(params,
NEATPopulation.PROPERTY_SURVIVAL_RATE));
result.setActivationCycles(EncogFileSection.parseInt(params,
NEATPopulation.PROPERTY_CYCLES));
}
}
// set factories
if (result.isHyperNEAT()) {
result.setGenomeFactory(new FactorHyperNEATGenome());
result.setCODEC(new HyperNEATCODEC());
} else {
result.setGenomeFactory(new FactorNEATGenome());
result.setCODEC(new NEATCODEC());
}
// set the next ID's
result.getInnovationIDGenerate().setCurrentID(nextInnovationID);
result.getGeneIDGenerate().setCurrentID(nextGeneID);
// find first genome, which should be the best genome
if (result.getSpecies().size() > 0) {
final Species species = result.getSpecies().get(0);
if (species.getMembers().size() > 0) {
result.setBestGenome(species.getMembers().get(0));
}
}
return result;
}
@Override
public void save(final OutputStream os, final Object obj) {
final EncogWriteHelper out = new EncogWriteHelper(os);
final NEATPopulation pop = (NEATPopulation) obj;
out.addSection("NEAT-POPULATION");
out.addSubSection("CONFIG");
out.writeProperty(PersistConst.ACTIVATION_CYCLES,
pop.getActivationCycles());
if (pop.isHyperNEAT()) {
out.writeProperty(NEATPopulation.PROPERTY_NEAT_ACTIVATION,
PersistNEATPopulation.TYPE_CPPN);
} else {
final ActivationFunction af = pop.getActivationFunctions()
.getList().get(0).getObj();
out.writeProperty(NEATPopulation.PROPERTY_NEAT_ACTIVATION, af);
}
out.writeProperty(PersistConst.INPUT_COUNT, pop.getInputCount());
out.writeProperty(PersistConst.OUTPUT_COUNT, pop.getOutputCount());
out.writeProperty(NEATPopulation.PROPERTY_CYCLES,
pop.getActivationCycles());
out.writeProperty(NEATPopulation.PROPERTY_POPULATION_SIZE,
pop.getPopulationSize());
out.writeProperty(NEATPopulation.PROPERTY_SURVIVAL_RATE,
pop.getSurvivalRate());
out.addSubSection("INNOVATIONS");
if (pop.getInnovations() != null) {
for (final String key : pop.getInnovations().getInnovations()
.keySet()) {
final NEATInnovation innovation = pop.getInnovations()
.getInnovations().get(key);
out.addColumn(key);
out.addColumn(innovation.getInnovationID());
out.addColumn(innovation.getNeuronID());
out.writeLine();
}
}
out.addSubSection("SPECIES");
// make sure the best species goes first
final Species bestSpecies = pop.determineBestSpecies();
if (bestSpecies != null) {
saveSpecies(out, bestSpecies);
}
// now write the other species, other than the best one
for (final Species species : pop.getSpecies()) {
if (species != bestSpecies) {
saveSpecies(out, species);
}
}
out.flush();
}
private void saveSpecies(final EncogWriteHelper out, final Species species) {
out.addColumn("s");
out.addColumn(species.getAge());
out.addColumn(species.getBestScore());
out.addColumn(species.getGensNoImprovement());
out.writeLine();
for (final Genome genome : species.getMembers()) {
final NEATGenome neatGenome = (NEATGenome) genome;
out.addColumn("g");
out.addColumn(neatGenome.getAdjustedScore());
out.addColumn(neatGenome.getScore());
out.addColumn(neatGenome.getBirthGeneration());
out.writeLine();
for (final NEATNeuronGene neatNeuronGene : neatGenome
.getNeuronsChromosome()) {
out.addColumn("n");
out.addColumn(neatNeuronGene.getId());
out.addColumn(neatNeuronGene.getActivationFunction());
out.addColumn(PersistNEATPopulation
.neuronTypeToString(neatNeuronGene.getNeuronType()));
out.addColumn(neatNeuronGene.getInnovationId());
out.writeLine();
}
for (final NEATLinkGene neatLinkGene : neatGenome
.getLinksChromosome()) {
out.addColumn("l");
out.addColumn(neatLinkGene.getId());
out.addColumn(neatLinkGene.isEnabled());
out.addColumn(neatLinkGene.getFromNeuronID());
out.addColumn(neatLinkGene.getToNeuronID());
out.addColumn(neatLinkGene.getWeight());
out.addColumn(neatLinkGene.getInnovationId());
out.writeLine();
}
}
}
}