/**
* After every {@link #interval} iteration, pick {@link #numberOfSentries a number of} random
* entities from the given {@link Algorithm algorithm's} topology and compare their previous
* fitness values with their current fitness values. An environment change is detected when the
* difference between the previous and current fitness values are >= the specified {@link
* #epsilon} value.
*
* @param algorithm used to get hold of topology of entities and number of iterations
* @return true if a change has been detected, false otherwise
*/
@Override
public <A extends HasTopology & Algorithm & HasNeighbourhood> boolean detect(A algorithm) {
if ((AbstractAlgorithm.get().getIterations() % interval == 0)
&& (AbstractAlgorithm.get().getIterations() != 0)) {
List all = Java.List_ArrayList().f(algorithm.getTopology());
for (int i = 0; i < numberOfSentries.getParameter(); i++) {
// select random sentry entity
int random = Rand.nextInt(all.size());
StandardParticle sentry = (StandardParticle) all.get(random);
// check for change
// double previousFitness = sentry.getFitness().getValue();
boolean detectedChange = false;
if (sentry.getFitness().getClass().getName().matches("MinimisationFitness")) {
Fitness previousFitness = sentry.getFitness();
sentry.calculateFitness();
Fitness currentFitness = sentry.getFitness();
if (Math.abs(previousFitness.getValue() - currentFitness.getValue()) >= epsilon) {
detectedChange = true;
}
} else if (sentry.getFitness().getClass().getName().matches("StandardMOFitness")) {
MOFitness previousFitness = (MOFitness) sentry.getFitness();
sentry.calculateFitness();
MOFitness currentFitness = (MOFitness) sentry.getFitness();
for (int k = 0; k < previousFitness.getDimension(); k++)
if (Math.abs(
previousFitness.getFitness(k).getValue()
- currentFitness.getFitness(k).getValue())
>= epsilon) {
detectedChange = true;
break;
}
}
if (detectedChange) {
System.out.println("Detected a change");
return true;
}
// remove the selected element from the all list preventing it from being selected again
all.remove(random);
}
}
return false;
}
/**
* Calculates the fitness of the given solution by setting the neural network weights to the
* solution and evaluating the training set in order to calculate the MSE (which is minimized).
*
* @param solution the weights representing a solution.
* @return a new MinimisationFitness wrapping the MSE training error.
*/
@Override
protected Fitness calculateFitness(Type solution) {
if (trainingSet == null) {
this.initialise();
}
int currentIteration = AbstractAlgorithm.get().getIterations();
if (currentIteration != previousShuffleIteration) {
try {
shuffler.operate(trainingSet);
} catch (CIlibIOException exception) {
exception.printStackTrace();
}
}
neuralNetwork.getArchitecture().accept(solutionConversionStrategy.interpretSolution(solution));
double errorTraining = 0.0;
OutputErrorVisitor visitor = new OutputErrorVisitor();
Vector error = null;
for (StandardPattern pattern : trainingSet) {
Vector output = neuralNetwork.evaluatePattern(pattern);
visitor.setInput(pattern);
neuralNetwork.getArchitecture().accept(visitor);
error = visitor.getOutput();
for (Numeric real : error) {
errorTraining += real.doubleValue() * real.doubleValue();
}
}
errorTraining /= trainingSet.getNumRows() * error.size();
return objective.evaluate(errorTraining);
}
/** {@inheritDoc} */
@Override
public Vector get(Particle particle) {
Vector localGuide = (Vector) particle.getLocalGuide();
Vector globalGuide = (Vector) particle.getGlobalGuide();
PSO pso = (PSO) AbstractAlgorithm.get();
List<Entity> positions = getRandomParentEntities(pso.getTopology());
// select three random individuals, all different and different from particle
ProbabilityDistributionFuction pdf = new UniformDistribution();
Vector position1 = (Vector) positions.get(0).getCandidateSolution();
Vector position2 = (Vector) positions.get(1).getCandidateSolution();
// Vector position3 = (Vector) positions.get(2).getContents();
Vector.Builder builder = Vector.newBuilder();
for (int i = 0; i < particle.getDimension(); ++i) {
double r = pdf.getRandomNumber(0, 1);
double attractor = r * localGuide.doubleValueOf(i) + (1 - r) * globalGuide.doubleValueOf(i);
double stepSize =
this.rand3.getRandomNumber(0, 1)
* (position1.doubleValueOf(i) - position2.doubleValueOf(i));
if (this.rand2.getRandomNumber(0, 1) > this.crossoverProbability.getParameter()) {
builder.add(attractor + stepSize);
} else {
builder.add(((Vector) particle.getPosition()).doubleValueOf(i)); // position3.getReal(i));
}
}
return builder.build();
}
/** {@inheritDoc} */
public void performReaction(PopulationBasedAlgorithm algorithm) {
// select new competitors and re-evaluate PBest vector of Particle
PopulationBasedAlgorithm currentAlgorithm =
(PopulationBasedAlgorithm) AbstractAlgorithm.get(); // the current sub population algorithm
int populationID = -1;
for (Entity e : currentAlgorithm.getTopology()) {
if (!(e instanceof AbstractParticle))
throw new RuntimeException(
"CompetitiveCoevolutionParticleReevaluationResponseStrategy should only be used with Particles");
if (populationID == -1)
populationID =
((Int) e.getProperties().get(EntityType.Coevolution.POPULATION_ID)).intValue();
Blackboard<Enum<?>, Type> blackboard = new Blackboard<Enum<?>, Type>();
blackboard.put(EntityType.CANDIDATE_SOLUTION, ((AbstractParticle) e).getBestPosition());
blackboard.put(EntityType.Coevolution.BOARD, new EntityScoreboard());
Fitness val = currentAlgorithm.getOptimisationProblem().getFitness(blackboard);
e.getProperties().put(EntityType.Particle.BEST_FITNESS, val);
// if currentV is better than re-evaluated pBest, then replace it
if (e.getFitness().compareTo(e.getBestFitness()) > 0) {
e.getProperties().put(EntityType.Particle.BEST_FITNESS, e.getFitness());
e.getProperties()
.put(
EntityType.Particle.BEST_POSITION,
e.getProperties().get(EntityType.CANDIDATE_SOLUTION).getClone());
}
}
}
@Override
public Vector get(Particle particle) {
Vector velocity = (Vector) particle.getVelocity();
Vector position = (Vector) particle.getPosition();
PSO algorithm = (PSO) AbstractAlgorithm.get();
int ns = (int) nSize.getParameter();
fj.data.List<Particle> neighbours =
algorithm
.getTopology()
.sort(
Ord.ord(
VectorBasedFunctions.sortByDistance(particle, new EuclideanDistanceMeasure())))
.take(ns);
Vector.Builder builder = Vector.newBuilder();
for (int i = 0; i < particle.getDimension(); ++i) {
double informationSum = 0.0;
double randomSum = 0;
for (Particle currentTarget : neighbours) {
Vector currentTargetPosition = (Vector) currentTarget.getBestPosition();
double randomComponent = Rand.nextDouble() * (4.1 / ns);
informationSum += randomComponent * currentTargetPosition.doubleValueOf(i);
randomSum += randomComponent;
}
double value =
inertiaWeight.getParameter()
* (velocity.doubleValueOf(i)
+ randomSum * ((informationSum / (ns * randomSum) - position.doubleValueOf(i))));
builder.add(value);
}
return builder.build();
}
@SuppressWarnings("unchecked")
@Override
public StructuredType get(Particle particle) {
MultiPopulationBasedAlgorithm topLevelAlgorithm =
(MultiPopulationBasedAlgorithm) AbstractAlgorithm.getAlgorithmList().get(0);
Blackboard<Enum<?>, Type> knowledge =
(Blackboard<Enum<?>, Type>)
this.knowledgeTransferStrategy.transferKnowledge(topLevelAlgorithm.getPopulations());
return (StructuredType) knowledge.get(EntityType.Particle.BEST_POSITION);
}
/** {@inheritDoc} */
@Override
public int compare(E o1, E o2) {
SinglePopulationBasedAlgorithm populationBasedAlgorithm =
(SinglePopulationBasedAlgorithm) AbstractAlgorithm.getAlgorithmList().index(0);
MOOptimisationProblem problem =
((MOOptimisationProblem) populationBasedAlgorithm.getOptimisationProblem());
Particle p1 = (Particle) o1;
Particle p2 = (Particle) o2;
MOFitness fitness1 = ((MOFitness) problem.getFitness(p1.getBestPosition()));
MOFitness fitness2 = ((MOFitness) problem.getFitness(p2.getBestPosition()));
int value = fitness1.compareTo(fitness2);
if (fitness1.compareTo(fitness2) == 0) {
int random = Rand.nextInt(20);
if (random > 10) value *= -1;
}
return value;
}
/**
* This is an Synchronous strategy:
*
* <ol>
* <li>For all particles:
* <ol>
* <li>Update the particle velocity
* <li>Update the particle position
* </ol>
* <li>For all particles:
* <ol>
* <li>Calculate the particle fitness
* <li>For all particles in the current particle's neighbourhood:
* <ol>
* <li>Update the neighbourhood best
* </ol>
* </ol>
* </ol>
*
* @see
* net.sourceforge.cilib.PSO.IterationStrategy#performIteration(net.sourceforge.cilib.PSO.PSO)
* @param pso The {@link PSO} to have an iteration applied.
*/
@Override
public void performIteration(PSO pso) {
Topology<Particle> topology = pso.getTopology();
for (Particle current : topology) {
current.updateVelocity();
current.updatePosition(); // TODO: replace with visitor (will simplify particle interface)
boundaryConstraint.enforce(current);
}
Problem problem = AbstractAlgorithm.getAlgorithmList().get(0).getOptimisationProblem();
for (Particle current : topology) {
current.calculateFitness();
for (Particle other : topology.neighbourhood(current)) {
Particle p1 = current.getNeighbourhoodBest().getClone();
Particle p2 = other.getNeighbourhoodBest().getClone();
OptimisationSolution s1 =
new OptimisationSolution(
p1.getCandidateSolution().getClone(),
problem.getFitness(p1.getCandidateSolution().getClone()));
OptimisationSolution s2 =
new OptimisationSolution(
p2.getCandidateSolution().getClone(),
problem.getFitness(p2.getCandidateSolution().getClone()));
MOFitness fitness1 = (MOFitness) s1.getFitness();
MOFitness fitness2 = (MOFitness) s2.getFitness();
// System.out.println("fitness1 = ");
// for (int i=0; i < fitness1.getDimension(); i++)
// System.out.println(fitness1.getFitness(i).getValue());
//
// System.out.println("fitness2 = ");
// for (int i=0; i < fitness2.getDimension(); i++)
// System.out.println(fitness2.getFitness(i).getValue());
if (fitness1.compareTo(fitness2) > 0) {
other.setNeighbourhoodBest(current);
}
}
}
}
/** Evaluates the function. */
@Override
public Double f(Vector x) {
this.tau = AbstractAlgorithm.get().getIterations();
return this.apply(this.tau, x);
}
/** {@inheritDoc} */
@Override
public Vector get(Particle particle) {
double averageParticleVelocity = 0.0;
Vector averageVelocity = null; // velocity.getClone();
// averageVelocity.reset();
PSO pso = (PSO) AbstractAlgorithm.get();
for (Particle p : pso.getTopology()) {
if (averageVelocity == null) {
averageVelocity = (Vector) p.getVelocity();
continue;
}
Vector particleVelocity = (Vector) p.getVelocity();
averageVelocity = averageVelocity.plus(particleVelocity);
averageParticleVelocity += particleVelocity.norm();
}
averageVelocity = averageVelocity.divide(particle.getDimension());
averageParticleVelocity /= particle.getDimension();
double swarmCenterVelocity = averageVelocity.norm();
double swarmCoherence = calculateSwarmCoherence(swarmCenterVelocity, averageParticleVelocity);
double sigmoidValue = this.sigmoid.apply(swarmCoherence);
Vector standardVelocity = this.delegate.get(particle);
Vector.Builder builder = Vector.newBuilder();
for (int i = 0; i < particle.getDimension(); ++i) {
double coherenceVelocity =
this.scalingFactor.getParameter()
* sigmoidValue
* averageVelocity.doubleValueOf(i)
* this.randomNumber.getRandomNumber();
builder.add(coherenceVelocity);
}
Vector coherence = builder.build();
return Vectors.sumOf(standardVelocity, coherence);
// float social = socialRandomGenerator.nextFloat();
// float cognitive = cognitiveRandomGenerator.nextFloat();
//
// //DistanceMeasure adm = new AbsoluteDistanceMeasure();
// //DistanceMeasure dm = new MetricDistanceMeasure();
//
// double avgv = 0.0;
// double swv = 0.0;
// Topology<Particle> topology = ((PSO)Algorithm.get()).getTopology();
// Iterator<? extends Particle> it = topology.neighbourhood(null);
// double[] al = new double[particle.getDimension()];
// while (it.hasNext()) {
// Particle pl = it.next();
// double tmpv = 0.0;
// //double tmpsv = 0.0;
// for(int dim = 0; dim < particle.getDimension(); dim++) {
// al[dim] = al[dim]+((Vector)pl.getVelocity()).getReal(dim);
// tmpv += Math.pow(((Vector)pl.getVelocity()).getReal(dim), 2);
// }
// tmpv = Math.sqrt(tmpv);
// avgv += tmpv;
// }
// for(int i = 0; i < particle.getDimension(); i++) {
// //al.set(i, ;
// swv += (al[i]/topology.size()) * (al[i]/topology.size());
// }
// swv = Math.sqrt(swv);
//
// for (int i = 0; i < particle.getDimension(); ++i) {
// double tmp = 0.0;
// tmp = inertiaWeight.getParameter()*velocity.getReal(i)
// + cognitive * (bestPosition.getReal(i) - position.getReal(i)) *
// cognitiveAcceleration.getParameter()
// + social * (nBestPosition.getReal(i) - position.getReal(i)) *
// socialAcceleration.getParameter();
//
// double avgdim = 0.0;
// it = topology.neighbourhood(null);
// while (it.hasNext()) {
// avgdim += ((Vector)(it.next().getVelocity())).getReal(i);
// }
// avgdim /= particle.getDimension();
//
// double cvelocity = MathUtil.sigmoid(swv/avgv)*avgdim*randomNumber.getCauchy();
//
// System.out.println(cvelocity);
// tmp += cvelocity;
//
// velocity.setReal(i, tmp);
//
// clamp(velocity, i);
// }
}
/** This is used to help with the unit test. i.e. no "empty stack exception" */
public int getIteration() {
return AbstractAlgorithm.get().getIterations();
}
/** Evaluates the function. */
@Override
public Double apply(Vector x) {
int iteration = AbstractAlgorithm.get().getIterations();
return this.apply(iteration, x);
}
