AckleyES.cpp

Go to the documentation of this file.

#include <shark/Algorithms/DirectSearch/Individual.h>
#include <shark/Algorithms/DirectSearch/Operators/Recombination/UniformCrossover.h>
#include <shark/Algorithms/DirectSearch/Operators/Recombination/SimulatedBinaryCrossover.h>
#include <shark/ObjectiveFunctions/Benchmarks/Ackley.h>
#include <shark/Statistics/Distributions/MultiVariateNormalDistribution.h>
 
using namespace shark;
 
namespace example {
 
    typedef Individual< RealVector, double, double> IndividualType;
    typedef std::vector< IndividualType > Population;
 
    struct FitnessComparator {
        bool operator()( const IndividualType & a, const IndividualType & b ) {
            return( a.unpenalizedFitness() < b.unpenalizedFitness() );
        }
 
    };
}
 
 
int main( int argc, char ** argv ) {
 
    const unsigned Mu           = 15;
    const unsigned Lambda       = 100;
    const unsigned Dimension    = 30;
    const double InitialSigma   = 3.;
    
    // Instantiate the objective function
    benchmarks::Ackley ackley( Dimension );
    
    // Initialize the mutation distribution
    MultiVariateNormalDistribution mutationDistribution;
    mutationDistribution.resize(Dimension);
    
    example::IndividualType prototypeIndividual;
    prototypeIndividual.chromosome() = InitialSigma;
    
    example::Population parents( Mu, prototypeIndividual );
    example::Population offspring( Lambda );
    
    // Initialize parents (not a god idea to start in a single point, shouldn't do this in practice)
    for(auto& ind: parents ) {
        ind.searchPoint() = ackley.proposeStartingPoint( );
    }
    
    // Evolutionary operators
    UniformCrossover uniform;
    
    // standard deviations for mutation of sigma
    double     tau0 = 1. / sqrt(2. * Dimension);
    double     tau1 = 1. / sqrt(2. * sqrt( static_cast<double>( Dimension ) ) );
 
    
    while( ackley.evaluationCounter() < 10000 ) {
    
        for( std::size_t i = 0; i < offspring.size(); i++ ) {
        
            // Select two parent individuals at random
            example::Population::const_iterator mom = parents.begin() + random::discrete(random::globalRng,  std::size_t(0), parents.size() - 1 );
            example::Population::const_iterator dad = parents.begin() + random::discrete(random::globalRng,  std::size_t(0), parents.size() - 1 );
        
            // Recombine step size
            offspring[i].chromosome() = random::uni(random::globalRng,  mom->chromosome(), dad->chromosome() );         
            // Mutate step size
            offspring[i].chromosome() *= random::logNormal(random::globalRng,  0, tau0 + tau1 );
            
            // Recombine search points
            offspring[i].searchPoint() = uniform(random::globalRng, mom->searchPoint(), dad->searchPoint() );
            // Mutate search point
            offspring[i].searchPoint() = offspring[i].chromosome() * mutationDistribution(random::globalRng).first;
    
            // Assign fitness
            offspring[i].unpenalizedFitness() = ackley.eval( offspring[i].searchPoint() );
        }
    
        // Selection 
        example::FitnessComparator comp;
        std::sort( offspring.begin(), offspring.end(), comp );
        std::copy( offspring.begin(), offspring.begin() + Mu, parents.begin() );
 
        
        std::cout   << ackley.evaluationCounter() << " " 
                << parents.front().unpenalizedFitness() << " "
                << parents.front().chromosome()  
                << std::endl;
    }   
}