/**
* Determine the fitness of the given Chromosome instance. The higher the return value, the more
* fit the instance. This method should always return the same fitness value for two equivalent
* Chromosome instances.
*
* @param a_subject the Chromosome instance to evaluate
* @return positive double reflecting the fitness rating of the given Chromosome
* @since 2.0 (until 1.1: return type int)
* @author Neil Rotstan, Klaus Meffert, John Serri
*/
public double evaluate(IChromosome a_subject) {
// Take care of the fitness evaluator. It could either be weighting higher
// fitness values higher (e.g.DefaultFitnessEvaluator). Or it could weight
// lower fitness values higher, because the fitness value is seen as a
// defect rate (e.g. DeltaFitnessEvaluator)
boolean defaultComparation = a_subject.getConfiguration().getFitnessEvaluator().isFitter(2, 1);
// The fitness value measures both how close the value is to the
// target amount supplied by the user and the total number of coins
// represented by the solution. We do this in two steps: first,
// we consider only the represented amount of change vs. the target
// amount of change and return higher fitness values for amounts
// closer to the target, and lower fitness values for amounts further
// away from the target. Then we go to step 2, which returns a higher
// fitness value for solutions representing fewer total coins, and
// lower fitness values for solutions representing more total coins.
// ------------------------------------------------------------------
int changeAmount = amountOfChange(a_subject);
int totalCoins = getTotalNumberOfCoins(a_subject);
int changeDifference = Math.abs(m_targetAmount - changeAmount);
double fitness;
if (defaultComparation) {
fitness = 0.0d;
} else {
fitness = MAX_BOUND / 2;
}
// Step 1: Determine distance of amount represented by solution from
// the target amount. If the change difference is greater than zero we
// will divide one by the difference in change between the
// solution amount and the target amount. That will give the desired
// effect of returning higher values for amounts closer to the target
// amount and lower values for amounts further away from the target
// amount.
// In the case where the change difference is zero it means that we have
// the correct amount and we assign a higher fitness value.
// ---------------------------------------------------------------------
if (defaultComparation) {
fitness += changeDifferenceBonus(MAX_BOUND / 2, changeDifference);
} else {
fitness -= changeDifferenceBonus(MAX_BOUND / 2, changeDifference);
}
// Step 2: We divide the fitness value by a penalty based on the number of
// coins. The higher the number of coins the higher the penalty and the
// smaller the fitness value.
// And inversely the smaller number of coins in the solution the higher
// the resulting fitness value.
// -----------------------------------------------------------------------
if (defaultComparation) {
fitness -= computeCoinNumberPenalty(MAX_BOUND / 2, totalCoins);
} else {
fitness += computeCoinNumberPenalty(MAX_BOUND / 2, totalCoins);
}
// Make sure fitness value is always positive.
// -------------------------------------------
return Math.max(1.0d, fitness);
}
private double computeFitness(IGPProgram a_program, Variable vx) {
double error = 0.0f;
Object[] noargs = new Object[0];
// Initialize local stores.
// ------------------------
a_program.getGPConfiguration().clearStack();
a_program.getGPConfiguration().clearMemory();
// Compute fitness for each program.
// ---------------------------------
for (int i = 2; i < 15; i++) {
for (int j = 0; j < a_program.size(); j++) {
vx.set(new Integer(i));
try {
try {
// Only evaluate after whole GP program was run.
// ---------------------------------------------
if (j == a_program.size() - 1) {
double result = a_program.execute_int(j, noargs);
error += Math.abs(result - fib_iter(i));
} else {
a_program.execute_void(j, noargs);
}
} catch (IllegalStateException iex) {
error = Double.MAX_VALUE / 2;
break;
}
} catch (ArithmeticException ex) {
System.out.println("Arithmetic Exception with x = " + i);
System.out.println(a_program.getChromosome(j));
throw ex;
}
}
}
return error;
}
protected void setRandomValue(double a_value) {
RandomGenerator randomGen = getGPConfiguration().getRandomGenerator();
m_value_double = randomGen.nextDouble() * (m_upperBounds - m_lowerBounds) + m_lowerBounds;
if (m_wholeNumbers) {
m_value_double = Math.round(m_value_double);
}
}
public void setValue(float a_value) {
if (m_wholeNumbers) {
m_value_float = Math.round(a_value);
} else {
m_value_float = a_value;
}
}
public void setValue(double a_value) {
if (m_wholeNumbers) {
m_value_double = Math.round(a_value);
} else {
m_value_double = a_value;
}
}
/**
* Determine the fitness of the given Chromosome instance. The higher the return value, the more
* fit the instance. This method should always return the same fitness value for two equivalent
* Chromosome instances.
*
* @param a_subject the Chromosome instance to evaluate
* @return positive double reflecting the fitness rating of the given Chromosome
* @since 2.0 (until 1.1: return type int)
* @author Neil Rotstan, Klaus Meffert, John Serri
*/
public double evaluate(IChromosome a_subject) {
int i = 0, j, k;
for (j = 0; j < EvolvingSubsumptionForRobocode.numberOfEvents; j++) {
eventPriority[j] = ((Integer) a_subject.getGene(i++).getAllele()).intValue();
}
for (j = 0; j < EvolvingSubsumptionForRobocode.numberOfBehaviours; j++) {
behaviourOverwrite[j] = ((Integer) a_subject.getGene(i++).getAllele()).intValue() > 50;
}
for (j = 0; j < EvolvingSubsumptionForRobocode.numberOfBehaviours; j++) {
for (k = 0; k < EvolvingSubsumptionForRobocode.behaviourSize; k++) {
behaviourActions[j][k] = ((Integer) a_subject.getGene(i++).getAllele()).intValue();
}
}
double tempFitness;
double fitness = 0;
for (j = 0; j < otherRobots.length; j++) {
RobotSpecification[] tempRobots = new RobotSpecification[2];
tempRobots[0] = evolvable;
tempRobots[1] = otherRobots[j];
battleSpecs = new BattleSpecification(1, new BattlefieldSpecification(800, 600), tempRobots);
tempFitness = 0;
System.out.println("Testing against " + otherRobots[j].getName() + "...");
for (i = 0; i < EvolvingSubsumptionForRobocode.numberOfBattles; i++) {
engine.runBattle(battleSpecs, true);
tempFitness += battleObserver.getScoreRobot() / 500;
}
tempFitness /= EvolvingSubsumptionForRobocode.numberOfBattles;
fitness += tempFitness;
}
fitness /= otherRobots.length;
return Math.min(1.0d, fitness);
}
protected void setRandomValue(float a_value) {
RandomGenerator randomGen = getGPConfiguration().getRandomGenerator();
m_value_float =
(float) (randomGen.nextFloat() * (m_upperBounds - m_lowerBounds) + m_lowerBounds);
if (m_wholeNumbers) {
m_value_float = Math.round(m_value_float);
}
}
/**
* Calculates the penalty to apply to the fitness value based on the ammount of coins in the
* solution
*
* @param a_maxFitness maximum fitness value allowed
* @param a_coins number of coins in the solution
* @return penalty for the fitness value base on the number of coins
* @author John Serri
* @since 2.2
*/
protected double computeCoinNumberPenalty(double a_maxFitness, int a_coins) {
if (a_coins == 1) {
// we know the solution cannot have less than one coin
return 0;
} else {
// The more coins the more penalty, but not more than the maximum fitness
// value possible. Let's avoid linear behavior and use
// exponential penalty calculation instead
return (Math.min(a_maxFitness, a_coins * a_coins));
}
}
/**
* Find and print the solution, return the solution error.
*
* @param a_conf the configuration to use
* @return absolute difference between the required and computed change
*/
protected int solve(
Configuration a_conf,
int a_targetChangeAmount,
SupergeneChangeFitnessFunction a_fitnessFunction,
Gene[] a_sampleGenes)
throws InvalidConfigurationException {
IChromosome sampleChromosome = new Chromosome(a_conf, a_sampleGenes);
a_conf.setSampleChromosome(sampleChromosome);
// Finally, we need to tell the Configuration object how many
// Chromosomes we want in our population. The more Chromosomes,
// the larger number of potential solutions (which is good for
// finding the answer), but the longer it will take to evolve
// the population (which could be seen as bad). We'll just set
// the population size to 500 here.
// ------------------------------------------------------------
a_conf.setPopulationSize(POPULATION_SIZE);
// Create random initial population of Chromosomes.
// ------------------------------------------------
Genotype population = Genotype.randomInitialGenotype(a_conf);
int s;
Evolution:
// Evolve the population, break if the the change solution is found.
// -----------------------------------------------------------------
for (int i = 0; i < MAX_ALLOWED_EVOLUTIONS; i++) {
population.evolve();
s =
Math.abs(
a_fitnessFunction.amountOfChange(population.getFittestChromosome())
- a_targetChangeAmount);
if (s == 0) {
break Evolution;
}
}
// Display the best solution we found.
// -----------------------------------
IChromosome bestSolutionSoFar = report(a_fitnessFunction, population);
return Math.abs(a_fitnessFunction.amountOfChange(bestSolutionSoFar) - a_targetChangeAmount);
}
/**
* Evolves the population of chromosomes within a genotype. This will execute all of the genetic
* operators added to the present active configuration and then invoke the natural selector to
* choose which chromosomes will be included in the next generation population.
*
* @param a_pop the population to evolve
* @param a_conf the configuration to use for evolution
* @return evolved population
* @author Klaus Meffert
* @since 3.2
*/
public Population evolve(Population a_pop, Configuration a_conf) {
Population pop = a_pop;
int originalPopSize = a_conf.getPopulationSize();
boolean monitorActive = a_conf.getMonitor() != null;
IChromosome fittest = null;
// If first generation: Set age to one to allow genetic operations,
// see CrossoverOperator for an illustration.
// ----------------------------------------------------------------
if (a_conf.getGenerationNr() == 0) {
int size = pop.size();
for (int i = 0; i < size; i++) {
IChromosome chrom = pop.getChromosome(i);
chrom.increaseAge();
}
} else {
// Select fittest chromosome in case it should be preserved and we are
// not in the very first generation.
// -------------------------------------------------------------------
if (a_conf.isPreserveFittestIndividual()) {
/** @todo utilize jobs. In pop do also utilize jobs, especially for fitness computation */
fittest = pop.determineFittestChromosome(0, pop.size() - 1);
}
}
if (a_conf.getGenerationNr() > 0) {
// Adjust population size to configured size (if wanted).
// Theoretically, this should be done at the end of this method.
// But for optimization issues it is not. If it is the last call to
// evolve() then the resulting population possibly contains more
// chromosomes than the wanted number. But this is no bad thing as
// more alternatives mean better chances having a fit candidate.
// If it is not the last call to evolve() then the next call will
// ensure the correct population size by calling keepPopSizeConstant.
// ------------------------------------------------------------------
keepPopSizeConstant(pop, a_conf);
}
// Ensure fitness value of all chromosomes is udpated.
// ---------------------------------------------------
if (monitorActive) {
// Monitor that fitness value of chromosomes is being updated.
// -----------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_BEFORE_UPDATE_CHROMOSOMES1,
a_conf.getGenerationNr(),
new Object[] {pop});
}
updateChromosomes(pop, a_conf);
if (monitorActive) {
// Monitor that fitness value of chromosomes is being updated.
// -----------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_AFTER_UPDATE_CHROMOSOMES1,
a_conf.getGenerationNr(),
new Object[] {pop});
}
// Apply certain NaturalSelectors before GeneticOperators will be executed.
// ------------------------------------------------------------------------
pop = applyNaturalSelectors(a_conf, pop, true);
// Execute all of the Genetic Operators.
// -------------------------------------
applyGeneticOperators(a_conf, pop);
// Reset fitness value of genetically operated chromosomes.
// Normally, this should not be necessary as the Chromosome class
// initializes each newly created chromosome with
// FitnessFunction.NO_FITNESS_VALUE. But who knows which Chromosome
// implementation is used...
// ----------------------------------------------------------------
int currentPopSize = pop.size();
for (int i = originalPopSize; i < currentPopSize; i++) {
IChromosome chrom = pop.getChromosome(i);
chrom.setFitnessValueDirectly(FitnessFunction.NO_FITNESS_VALUE);
// Mark chromosome as new-born.
// ----------------------------
chrom.resetAge();
// Mark chromosome as being operated on.
// -------------------------------------
chrom.increaseOperatedOn();
}
// Increase age of all chromosomes which are not modified by genetic
// operations.
// -----------------------------------------------------------------
int size = Math.min(originalPopSize, currentPopSize);
for (int i = 0; i < size; i++) {
IChromosome chrom = pop.getChromosome(i);
chrom.increaseAge();
// Mark chromosome as not being operated on.
// -----------------------------------------
chrom.resetOperatedOn();
}
// If a bulk fitness function has been provided, call it.
// ------------------------------------------------------
BulkFitnessFunction bulkFunction = a_conf.getBulkFitnessFunction();
if (bulkFunction != null) {
if (monitorActive) {
// Monitor that bulk fitness will be called for evaluation.
// --------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_BEFORE_BULK_EVAL,
a_conf.getGenerationNr(),
new Object[] {bulkFunction, pop});
}
/** @todo utilize jobs: bulk fitness function is not so important for a prototype! */
bulkFunction.evaluate(pop);
if (monitorActive) {
// Monitor that bulk fitness has been called for evaluation.
// ---------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_AFTER_BULK_EVAL,
a_conf.getGenerationNr(),
new Object[] {bulkFunction, pop});
}
}
// Ensure fitness value of all chromosomes is udpated.
// ---------------------------------------------------
if (monitorActive) {
// Monitor that fitness value of chromosomes is being updated.
// -----------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_BEFORE_UPDATE_CHROMOSOMES2,
a_conf.getGenerationNr(),
new Object[] {pop});
}
updateChromosomes(pop, a_conf);
if (monitorActive) {
// Monitor that fitness value of chromosomes is being updated.
// -----------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_AFTER_UPDATE_CHROMOSOMES2,
a_conf.getGenerationNr(),
new Object[] {pop});
}
// Apply certain NaturalSelectors after GeneticOperators have been applied.
// ------------------------------------------------------------------------
pop = applyNaturalSelectors(a_conf, pop, false);
// Fill up population randomly if size dropped below specified percentage
// of original size.
// ----------------------------------------------------------------------
if (a_conf.getMinimumPopSizePercent() > 0) {
int sizeWanted = a_conf.getPopulationSize();
int popSize;
int minSize = (int) Math.round(sizeWanted * (double) a_conf.getMinimumPopSizePercent() / 100);
popSize = pop.size();
if (popSize < minSize) {
IChromosome newChrom;
IChromosome sampleChrom = a_conf.getSampleChromosome();
Class sampleChromClass = sampleChrom.getClass();
IInitializer chromIniter =
a_conf.getJGAPFactory().getInitializerFor(sampleChrom, sampleChromClass);
while (pop.size() < minSize) {
try {
/**
* @todo utilize jobs as initialization may be time-consuming as invalid combinations
* may have to be filtered out
*/
newChrom = (IChromosome) chromIniter.perform(sampleChrom, sampleChromClass, null);
if (monitorActive) {
// Monitor that fitness value of chromosomes is being updated.
// -----------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_BEFORE_ADD_CHROMOSOME,
a_conf.getGenerationNr(),
new Object[] {pop, newChrom});
}
pop.addChromosome(newChrom);
} catch (Exception ex) {
throw new RuntimeException(ex);
}
}
}
}
IChromosome newFittest = reAddFittest(pop, fittest);
if (monitorActive && newFittest != null) {
// Monitor that fitness value of chromosomes is being updated.
// -----------------------------------------------------------
a_conf
.getMonitor()
.event(
IEvolutionMonitor.MONITOR_EVENT_READD_FITTEST,
a_conf.getGenerationNr(),
new Object[] {pop, fittest});
}
// Increase number of generations.
// -------------------------------
a_conf.incrementGenerationNr();
// Fire an event to indicate we've performed an evolution.
// -------------------------------------------------------
m_lastPop = pop;
m_lastConf = a_conf;
a_conf
.getEventManager()
.fireGeneticEvent(new GeneticEvent(GeneticEvent.GENOTYPE_EVOLVED_EVENT, this));
return pop;
}
public void operate(final Population a_population, final List a_candidateChromosomes) {
// Work out the number of crossovers that should be performed.
// -----------------------------------------------------------
int m_crossoverRate = getCrossOverRate();
double m_crossoverRatePercent = getCrossOverRatePercent();
int size = Math.min(getConfiguration().getPopulationSize(), a_population.size());
int numCrossovers = 0;
if (m_crossoverRate >= 0) {
numCrossovers = size / m_crossoverRate;
} else if (m_crossoverRateCalc != null) {
numCrossovers = size / 6;
} else {
numCrossovers = (int) (size * m_crossoverRatePercent);
}
RandomGenerator generator = getConfiguration().getRandomGenerator();
IGeneticOperatorConstraint constraint =
getConfiguration().getJGAPFactory().getGeneticOperatorConstraint();
// For each crossover, grab two random chromosomes, pick a random
// locus (gene location), and then swap that gene and all genes
// to the "right" (those with greater loci) of that gene between
// the two chromosomes.
// --------------------------------------------------------------
int index1, index2;
crossoverValues = new int[size];
for (int i = 0; i < size; i++) {
crossoverValues[i] = 0;
}
List<Integer> chromeSel = new ArrayList<Integer>();
for (int i = 0; i < size; i++) {
if (m_crossoverRateCalc != null) {
CrossoverRateCalculator ccalc = (CrossoverRateCalculator) m_crossoverRateCalc;
ccalc.setCurrentFitness(a_population.getChromosome(i).getFitnessValue());
int numbers = ccalc.calculateCurrentRate();
crossoverValues[i] = numbers;
while (numbers > 0) {
chromeSel.add(i);
numbers--;
}
}
}
int justtocheck = chromeSel.size();
if (m_crossoverRateCalc != null) {
if (justtocheck == 0) {
System.out.println("Exception Caught.. Crossover.. chromesel array is null!!");
for (int fix = 0; fix <= size - 2; fix++) {
chromeSel.add(fix);
}
}
}
//
for (int i = 0; i < numCrossovers; i++) {
IChromosome chrom1;
IChromosome chrom2;
if (m_crossoverRateCalc != null) {
index1 = generator.nextInt(chromeSel.size());
index2 = generator.nextInt(chromeSel.size());
chrom1 = a_population.getChromosome(chromeSel.get(index1));
chrom2 = a_population.getChromosome(chromeSel.get(index2));
} else {
index1 = generator.nextInt(size);
index2 = generator.nextInt(size);
chrom1 = a_population.getChromosome(index1);
chrom2 = a_population.getChromosome(index2);
}
// Verify that crossover is allowed.
// ---------------------------------
if (!isXoverNewAge() && chrom1.getAge() < 1 && chrom2.getAge() < 1) {
// Crossing over two newly created chromosomes is not seen as helpful
// here.
// ------------------------------------------------------------------
continue;
}
if (constraint != null) {
List v = new Vector();
v.add(chrom1);
v.add(chrom2);
if (!constraint.isValid(a_population, v, this)) {
// Constraint forbids crossing over.
// ---------------------------------
continue;
}
}
// Clone the chromosomes.
// ----------------------
IChromosome firstMate = (IChromosome) ((ICloneable) chrom1).clone();
IChromosome secondMate = (IChromosome) ((ICloneable) chrom2).clone();
// Cross over the chromosomes.
// ---------------------------
doCrossover(firstMate, secondMate, a_candidateChromosomes, generator);
}
}
/**
* Calculate Cumulative Cauchy distribution function.
*
* @return the probability that a stochastic variable x is less than X
* @author Klaus Meffert
* @since 1.1
*/
public double nextCauchy() {
return 0.5 + Math.atan((m_rn.nextDouble() - m_location) / m_scale) / Math.PI;
}
public int nextInt(final int a_ceiling) {
return Math.min(a_ceiling - 1, (int) Math.round(nextCauchy() * a_ceiling));
}
public int nextInt() {
return Math.min(Integer.MAX_VALUE - 1, (int) Math.round(nextCauchy() * Integer.MAX_VALUE));
}
public CommandGene applyMutation(int index, double a_percentage)
throws InvalidConfigurationException {
// If percentage is very high: do mutation not relying on
// current value but on a random value.
// ------------------------------------------------------
if (a_percentage > 0.85d) {
setRandomValue();
} else {
Class retType = getReturnType();
if (retType == CommandGene.FloatClass) {
float newValuef;
float rangef = ((float) m_upperBounds - (float) m_lowerBounds) * (float) a_percentage;
if (m_value_float >= (m_upperBounds - m_lowerBounds) / 2) {
newValuef =
m_value_float - getGPConfiguration().getRandomGenerator().nextFloat() * rangef;
} else {
newValuef =
m_value_float + getGPConfiguration().getRandomGenerator().nextFloat() * rangef;
}
// Ensure value is within bounds.
// ------------------------------
if (m_lowerBounds - newValuef > DELTA || newValuef - m_upperBounds > DELTA) {
setRandomValue(m_value_float);
} else {
setValue(newValuef);
}
} else if (retType == CommandGene.DoubleClass) {
double newValueD;
double rangeD = (m_upperBounds - m_lowerBounds) * a_percentage;
if (m_value_double >= (m_upperBounds - m_lowerBounds) / 2) {
newValueD =
m_value_double - getGPConfiguration().getRandomGenerator().nextFloat() * rangeD;
} else {
newValueD =
m_value_double + getGPConfiguration().getRandomGenerator().nextFloat() * rangeD;
}
// Ensure value is within bounds.
// ------------------------------
if (m_lowerBounds - newValueD > DELTA || newValueD - m_upperBounds > DELTA) {
setRandomValue(m_value_float);
} else {
setValue(newValueD);
}
} else if (retType == CommandGene.IntegerClass) {
int newValueI;
double range = (m_upperBounds - m_lowerBounds) * a_percentage;
if (m_value_int >= (m_upperBounds - m_lowerBounds) / 2) {
newValueI =
m_value_int
- (int) Math.round(getGPConfiguration().getRandomGenerator().nextInt() * range);
} else {
newValueI =
m_value_int
+ (int) Math.round(getGPConfiguration().getRandomGenerator().nextFloat() * range);
}
// Ensure value is within bounds.
// ------------------------------
if (newValueI < m_lowerBounds || newValueI > m_upperBounds) {
setRandomValue(m_value_int);
} else {
setValue(newValueI);
}
} else if (retType == CommandGene.LongClass) {
long newValueL;
double range = (m_upperBounds - m_lowerBounds) * a_percentage;
if (m_value_long >= (m_upperBounds - m_lowerBounds) / 2) {
newValueL =
m_value_long
- Math.round(getGPConfiguration().getRandomGenerator().nextInt() * range);
} else {
newValueL =
m_value_long
+ Math.round(getGPConfiguration().getRandomGenerator().nextFloat() * range);
}
// Ensure value is within bounds.
// ------------------------------
if (newValueL < m_lowerBounds || newValueL > m_upperBounds) {
setRandomValue(m_value_long);
} else {
setValue(newValueL);
}
}
}
return this;
}
public long nextLong() {
return Math.min(Long.MAX_VALUE - 1, Math.round(nextCauchy() * Long.MAX_VALUE));
}
public float nextFloat() {
return Math.min(Float.MAX_VALUE - 1, (float) (nextCauchy() * Float.MAX_VALUE));
}
/**
* Executes the genetic algorithm to determine the minimum number of items necessary to make up
* the given target volume. The solution will then be written to the console.
*
* @param a_knapsackVolume the target volume for which this method is attempting to produce the
* optimal list of items
* @throws Exception
* @author Klaus Meffert
* @since 2.3
*/
public static void findItemsForVolume(double a_knapsackVolume) throws Exception {
// Start with a DefaultConfiguration, which comes setup with the
// most common settings.
// -------------------------------------------------------------
Configuration conf = new DefaultConfiguration();
conf.setPreservFittestIndividual(true);
// Set the fitness function we want to use. We construct it with
// the target volume passed in to this method.
// ---------------------------------------------------------
FitnessFunction myFunc = new KnapsackFitnessFunction(a_knapsackVolume);
conf.setFitnessFunction(myFunc);
// Now we need to tell the Configuration object how we want our
// Chromosomes to be setup. We do that by actually creating a
// sample Chromosome and then setting it on the Configuration
// object. As mentioned earlier, we want our Chromosomes to each
// have as many genes as there are different items available. We want the
// values (alleles) of those genes to be integers, which represent
// how many items of that type we have. We therefore use the
// IntegerGene class to represent each of the genes. That class
// also lets us specify a lower and upper bound, which we set
// to senseful values (i.e. maximum possible) for each item type.
// --------------------------------------------------------------
Gene[] sampleGenes = new Gene[itemVolumes.length];
for (int i = 0; i < itemVolumes.length; i++) {
sampleGenes[i] = new IntegerGene(conf, 0, (int) Math.ceil(a_knapsackVolume / itemVolumes[i]));
}
IChromosome sampleChromosome = new Chromosome(conf, sampleGenes);
conf.setSampleChromosome(sampleChromosome);
// Finally, we need to tell the Configuration object how many
// Chromosomes we want in our population. The more Chromosomes,
// the larger number of potential solutions (which is good for
// finding the answer), but the longer it will take to evolve
// the population (which could be seen as bad).
// ------------------------------------------------------------
conf.setPopulationSize(50);
// Create random initial population of Chromosomes.
// Here we try to read in a previous run via XMLManager.readFile(..)
// for demonstration purpose!
// -----------------------------------------------------------------
Genotype population;
try {
Document doc = XMLManager.readFile(new File("knapsackJGAP.xml"));
population = XMLManager.getGenotypeFromDocument(conf, doc);
} catch (FileNotFoundException fex) {
population = Genotype.randomInitialGenotype(conf);
}
population = Genotype.randomInitialGenotype(conf);
// Evolve the population. Since we don't know what the best answer
// is going to be, we just evolve the max number of times.
// ---------------------------------------------------------------
for (int i = 0; i < MAX_ALLOWED_EVOLUTIONS; i++) {
population.evolve();
}
// Save progress to file. A new run of this example will then be able to
// resume where it stopped before!
// ---------------------------------------------------------------------
// represent Genotype as tree with elements Chromomes and Genes
// ------------------------------------------------------------
DataTreeBuilder builder = DataTreeBuilder.getInstance();
IDataCreators doc2 = builder.representGenotypeAsDocument(population);
// create XML document from generated tree
// ---------------------------------------
XMLDocumentBuilder docbuilder = new XMLDocumentBuilder();
Document xmlDoc = (Document) docbuilder.buildDocument(doc2);
XMLManager.writeFile(xmlDoc, new File("knapsackJGAP.xml"));
// Display the best solution we found.
// -----------------------------------
IChromosome bestSolutionSoFar = population.getFittestChromosome();
System.out.println(
"The best solution has a fitness value of " + bestSolutionSoFar.getFitnessValue());
System.out.println("It contained the following: ");
int count;
double totalVolume = 0.0d;
for (int i = 0; i < bestSolutionSoFar.size(); i++) {
count = ((Integer) bestSolutionSoFar.getGene(i).getAllele()).intValue();
if (count > 0) {
System.out.println("\t " + count + " x " + itemNames[i]);
totalVolume += itemVolumes[i] * count;
}
}
System.out.println("\nFor a total volume of " + totalVolume + " ccm");
System.out.println("Expected volume was " + a_knapsackVolume + " ccm");
System.out.println("Volume difference is " + Math.abs(totalVolume - a_knapsackVolume) + " ccm");
}
protected void setRandomValue(long a_value) {
RandomGenerator randomGen = getGPConfiguration().getRandomGenerator();
m_value_long =
Math.round(randomGen.nextDouble() * (m_upperBounds - m_lowerBounds) + m_lowerBounds);
}
