praktikum-holons/src/algorithm/topologie/PsoAlgorithm.java

package algorithm.topologie;

import java.util.ArrayList;
import java.util.List;
import java.util.ListIterator;
import java.util.TreeSet;
import java.util.stream.Collectors;

import algorithm.objectiveFunction.ObjectiveFunctionByCarlos;
import algorithm.objectiveFunction.TopologieObjectiveFunction;
import api.TopologieAlgorithmFramework;
import api.AlgorithmFrameworkFlex.Individual;
import ui.model.DecoratedState;

public class PsoAlgorithm extends TopologieAlgorithmFramework {

	private int popsize = 20;
	private int maxGenerations = 100;
	private double dependency = 2.07; 
	private double c1, c2, w;
	private double maxVelocity = 4.0;
	private double deviation = 0.5;
	//mutation
	private int mutationInterval = 1;
	private boolean useIntervalMutation = true;
	private double mutationRate = 0.01;
	private double mutateProbabilityInterval = 0.01;
	private double maxMutationPercent = 0.01;
	
	private boolean moreInformation = false;
	
	public PsoAlgorithm(){
		addIntParameter("popsize", popsize, intValue -> popsize = intValue, () -> popsize, 1);
		addIntParameter("maxGenerations", maxGenerations, intValue -> maxGenerations = intValue, () -> maxGenerations, 1);
		addDoubleParameter("deviation", deviation, doubleValue -> deviation = doubleValue, () -> deviation, 0);
		addDoubleParameter("dependency", dependency, doubleValue -> dependency = doubleValue, () -> dependency, 2.001, 2.4);
		addIntParameter("mutationInterval", mutationInterval, intValue -> mutationInterval = intValue, () -> mutationInterval, 0);
		addBooleanParameter("useIntervalMutation", useIntervalMutation, booleanValue -> useIntervalMutation = booleanValue);
		addDoubleParameter("mutateProbabilityInterval", mutateProbabilityInterval, doubleValue -> mutateProbabilityInterval = doubleValue, () -> mutateProbabilityInterval, 0.0, 1.0);
		addDoubleParameter("mutationRate", mutationRate, doubleValue -> mutationRate = doubleValue, () -> mutationRate, 0.0, 1.0);
		addDoubleParameter("maxMutationPercent", maxMutationPercent, doubleValue -> maxMutationPercent = doubleValue, () -> maxMutationPercent, 0.0, 1.0);
		addDoubleParameter("maxVelocity", maxVelocity, doubleValue -> maxVelocity = doubleValue, () -> maxVelocity, 0.0);
		addBooleanParameter("moreInformation", moreInformation, booleanValue -> moreInformation = booleanValue);

	}
	
	
	public static double linearInterpolate(double first, double second, double alpha) {
		return first * (1.0 - alpha) + second * alpha;
	}
	public static double inverseLinearInterpolation(double min, double max, double value) {
		if (max - min == 0) return max;
		else return (value - min) / (max - min);
	}
	
	@Override
	protected double evaluateState(DecoratedState actualstate) {
		return TopologieObjectiveFunction.getFitnessValueForState(actualstate);
	}

	@Override
	/**
	 *  <p>Algo from Paper:</p><font size="3"><pre>
	 *  
	 *  Begin
	 *  	t = 0; {t: generation index}
	 *  	initialize particles x<sub>p,i,j</sub>(t);
	 *  	evaluation x<sub>p,i,j</sub>(t);
	 *  	while (termination condition &ne; true) do
	 *  		v<sub>i,j</sub>(t) = update v<sub>i,j</sub>(t); {by Eq. (6)}
	 *  		x<sub>g,i,j</sub>(t) = update x<sub>g,i,j</sub>(t); {by Eq. (7)}
	 *  		x<sub>g,i,j</sub>(t) = mutation x<sub>g,i,j</sub>(t); {by Eq. (11)}
	 *  		x<sub>p,i,j</sub>(t) = decode x<sub>g,i,j</sub>(t); {by Eqs. (8) and (9)}
	 *  		evaluate x<sub>p,i,j</sub>(t);
	 *  		t = t + 1;
	 *  	end while
	 *  End</pre></font>
	 *  <p>with:</p><font size="3">
	 *  
	 *  x<sub>g,i,j</sub>: genotype ->genetic information -> in continuous space<br>
	 *  x<sub>p,i,j</sub>: phenotype -> observable characteristics-> in binary space<br>
	 *  X<sub>g,max</sub>: is the Maximum here set to 4.<br>
	 *  Eq. (6):v<sub>i,j</sub>(t + 1) = wv<sub>i,j</sub>+c<sub>1</sub>R<sub>1</sub>(P<sub>best,i,j</sub>-x<sub>p,i,j</sub>(t))+c<sub>2</sub>R<sub>2</sub>(g<sub>best,i,j</sub>-x<sub>p,i,j</sub>(t))<br>
	 *  Eq. (7):x<sub>g,i,j</sub>(t + 1) = x<sub>g,i,j</sub>(t) + v<sub>i,j</sub>(t + 1)<br>
	 *  Eq. (11):<b>if(</b>rand()&lt;r<sub>mu</sub><b>)then</b> x<sub>g,i,j</sub>(t + 1) = -x<sub>g,i,j</sub>(t + 1)<br>
	 *  Eq. (8):x<sub>p,i,j</sub>(t + 1) = <b>(</b>rand() &lt; S(x<sub>g,i,j</sub>(t + 1))<b>) ?</b> 1 <b>:</b> 0<br>
	 *  Eq. (9) Sigmoid:S(x<sub>g,i,j</sub>(t + 1)) := 1/(1 + e<sup>-x<sub>g,i,j</sub>(t + 1)</sup>)<br></font>
	 *  <p>Parameter:</p>
	 *  w inertia, calculated from phi(Variable:{@link #dependency})<br>
	 *  c1:	influence, calculated from phi(Variable:{@link #dependency}) <br>
	 *  c2:	influence, calculated from phi(Variable:{@link #dependency})<br>
	 *  r<sub>mu</sub>: probability that the proposed operation is conducted defined by limit(Variable:{@link #limit})<br>
	 *  
	 *  
	 */
	protected Individual executeAlgo() {
		resetWildcards();
		initDependentParameter();
		Individual globalBest = new Individual();
		globalBest.position = extractPositionAndAccess();
		globalBest.fitness = evaluatePosition(globalBest.position);
		console.println("Start Value:" + globalBest.fitness);
		int dimensions = globalBest.position.size();
		List<Particle> swarm= initializeParticles(dimensions);
		List<Double> runList = new ArrayList<Double>();
		runList.add(globalBest.fitness);
		evaluation(globalBest, swarm);
		runList.add(globalBest.fitness);
		for (int iteration = 0; iteration < this.maxGenerations ; iteration++) {
			int mutationAllowed = iteration % mutationInterval;
			double bitsFlipped = 0;
			for (int particleNumber = 0; particleNumber < this.popsize; particleNumber++) {
				Particle particle = swarm.get(particleNumber);		
				
				if(this.useIntervalMutation) {
					boolean allowMutation = (Random.nextDouble() <  this.mutateProbabilityInterval);
					TreeSet<Integer> mutationLocationSet = null;
					if(allowMutation)mutationLocationSet = locationsToMutate(dimensions);
					for(int index = 0; index < dimensions; index++) {
						updateVelocity(particle, index, globalBest);
						updateGenotype(particle, index);
						if(allowMutation &&mutationAllowed == 0 && iteration != 0 && mutationLocationSet.contains(index))mutation(particle, index);
						decode(particle, index);
					}
				}else {				
					int count = 0;
					for(int index = 0; index < dimensions; index++) {
						updateVelocity(particle, index, globalBest);
						updateGenotype(particle, index);
						if(mutationAllowed == 0 && iteration != 0 && Random.nextDouble() < mutationRate) {
							count++;
							mutation(particle, index);
						}
						decode(particle, index);
					}
					bitsFlipped += count;
				}
			}
			if(moreInformation) console.println("\t\t\t\t\t\tAvgBitsMutate: " + (bitsFlipped / (double)popsize));
			if(cancel)return null;
			evaluation(globalBest, swarm);
			runList.add(globalBest.fitness);
			if(moreInformation) console.println("------------------------");
		}
		console.println(" End Value:" + globalBest.fitness);
		this.runList = runList;
		return globalBest;
	}

	@Override
	protected int getProgressBarMaxCount() {
		return rounds * maxGenerations * popsize + 1;
	}

	@Override
	protected String algoInformationToPrint() {
		return "GA for topologie generation";
	}

	@Override
	protected String plottFileName() {
		return "ga-topologie.txt";
	}
	
	
	/**
	 * 
	 * @param problemSize maximum index of position in the particle
	 * @return
	 */
	private List<Particle> initializeParticles(int problemSize) {
		List<Particle> swarm = new ArrayList<Particle>();
		//Create The Particle
		for (int particleNumber = 0; particleNumber < popsize; particleNumber++){
			//Create a Random position
			List<Integer> aRandomPosition = new ArrayList<Integer>();
			for (int index = 0; index < problemSize; index++){
				aRandomPosition.add(Random.nextIntegerInRange(0, this.getMaximumIndexObjects(index) + 1));
			}
			swarm.add(new Particle(aRandomPosition));
		}
		return swarm;
	}
	/**
	 * Calculate w, c1, c2
	 */
	private void initDependentParameter() {
		w = 1.0 / (dependency - 1 + Math.sqrt(dependency * dependency - 2 * dependency));
		c1 = c2 = dependency * w;
	}
	/**
	 * Evaluate each particle and update the global Best position;
	 * @param globalBest
	 * @param swarm
	 */
	private void evaluation(Individual globalBest, List<Particle> swarm) {
		for(Particle p: swarm) {
			double localEvaluationValue = evaluatePosition(p.xPhenotype);
			if(moreInformation) console.println("Fitness " + localEvaluationValue);
			p.checkNewEvaluationValue(localEvaluationValue);
			if(localEvaluationValue < globalBest.fitness) {
				globalBest.fitness = localEvaluationValue;
				globalBest.position = p.localBest.position;
			}
		}
	}
	/**
	 * 	Eq. (6):v<sub>i,j</sub>(t + 1) = wv<sub>i,j</sub>+c<sub>1</sub>R<sub>1</sub>(P<sub>best,i,j</sub>-x<sub>p,i,j</sub>(t))+c<sub>2</sub>R<sub>2</sub>(g<sub>best,i,j</sub>-x<sub>p,i,j</sub>(t))<br>
	 * @param particle
	 * @param index
	 * @param globalBest
	 */
	private void updateVelocity(Particle particle, int index, Individual globalBest) {
		double r1 = Random.nextDouble();
		double r2 =	Random.nextDouble();
		double posValue = particle.xPhenotype.get(index);
		particle.velocity.set(index, clamp(w*particle.velocity.get(index) + c1*r1*((particle.localBest.position.get(index))  - posValue) + c2*r2*((globalBest.position.get(index))- posValue)) );
	}
	/**
	 * Eq. (7):x<sub>g,i,j</sub>(t + 1) = x<sub>g,i,j</sub>(t) + v<sub>i,j</sub>(t + 1)<br>
	 * @param particle
	 * @param index
	 */
	private void updateGenotype(Particle particle, int index) {
		particle.xGenotype.set(index, clamp(particle.xGenotype.get(index) + particle.velocity.get(index)));
	}
	/**
	 * Eq. (11):<b>if(</b>rand()&lt;r<sub>mu</sub><b>)then</b> x<sub>g,i,j</sub>(t + 1) = -x<sub>g,i,j</sub>(t + 1)<br>
	 * @param particle
	 * @param index
	 */
	private void mutation(Particle particle, int index) {
		//if(Random.nextDouble() < limit) 
			particle.xGenotype.set(index, -particle.xGenotype.get(index));
	}
	/**
	 * Eq. (8):x<sub>p,i,j</sub>(t + 1) = <b>(</b>rand() &lt; S(x<sub>g,i,j</sub>(t + 1))<b>) ?</b> 1 <b>:</b> 0<br>
	 * @param particle
	 * @param index
	 */
	private void decode(Particle particle, int index) {
		double value = clamp(Random.nextGaussian(particle.xGenotype.get(index), deviation));
		double alpha = inverseLinearInterpolation(-maxVelocity, +maxVelocity, value);
		double result = linearInterpolate(0, this.getMaximumIndexObjects(index), alpha);
		particle.xPhenotype.set(index, (int)Math.round(result));
	}
	/**
	 * Eq. (9) Sigmoid:S(x<sub>g,i,j</sub>(t + 1)) := 1/(1 + e<sup>-x<sub>g,i,j</sub>(t + 1)</sup>)<br></font>
	 * @param value
	 * @return
	 */
	private double Sigmoid(double value) {
		return 1.0 / (1.0 + Math.exp(-value));
	}

	/**
	 * To clamp X<sub>g,j,i</sub> and v<sub>i,j</sub> in Range [-X<sub>g,max</sub>|+X<sub>g,max</sub>] with {X<sub>g,max</sub>= 4}
	 * @param value
	 * @return
	 */
	private double clamp(double value) {
		return Math.max(-maxVelocity, Math.min(maxVelocity, value));
	}
	private TreeSet<Integer> locationsToMutate(int dimensions) {
		TreeSet<Integer> mutationLocation = new TreeSet<Integer>(); //sortedSet
		int maximumAmountOfMutatedBits = Math.max(1, (int)Math.round(((double) dimensions) * this.maxMutationPercent));
		int randomUniformAmountOfMutatedValues = Random.nextIntegerInRange(1,maximumAmountOfMutatedBits + 1);
		for(int i = 0; i< randomUniformAmountOfMutatedValues; i++) {
			boolean success = mutationLocation.add(Random.nextIntegerInRange(0, dimensions));
			if(!success) i--; //can be add up to some series long loops if maximumAmountOfMutatedBits get closed to problemsize.
		}
		return mutationLocation;
	}
	
	
	
	
	
	
	/**
	 * Class to represent a Particle.
	 */
	private class Particle{
		/**
		 * The velocity of a particle.
		 */
		public List<Double> velocity;
		/**
		 * The positions genotype.
		 */
		public List<Double> xGenotype;
		/**
		 * The positions phenotype. Alias the current position.
		 */
		public List<Integer> xPhenotype;
		
		public Individual localBest;
		
		Particle(List<Integer> position){
			this.xPhenotype = position;
			//Init velocity, xGenotype with 0.0 values.
			this.velocity = position.stream().map(bool -> 0.0).collect(Collectors.toList());
			this.xGenotype = position.stream().map(bool -> 0.0).collect(Collectors.toList());
			localBest = new Individual();
			localBest.fitness = Double.MAX_VALUE;
		}
		public void checkNewEvaluationValue(double newEvaluationValue) {
			if(newEvaluationValue < localBest.fitness) {
				localBest.fitness = newEvaluationValue;
				localBest.position = xPhenotype.stream().collect(Collectors.toList());
			}
		}
		public String toString() {
			return "Particle with xPhenotype(Position), xGenotype, velocity:[" 
					+ listToString(xPhenotype) + "],[" + listToString(xGenotype) + "],[" 
					+ listToString(velocity) + "]";
		}
		private <Type> String listToString(List<Type> list) {
			return list.stream().map(Object::toString).collect(Collectors.joining(", "));
		}
		
	}
	
}