AlgCuckooSearch.cpp

#include "ECF_base.h"
#include "floatingpoint/FloatingPoint.h"
#include "AlgCuckooSearch.h"
#include <boost/random/normal_distribution.hpp>
#include <boost/random.hpp>
#include <ctime>
#include <cstdlib>
#include <vector>
 
 
CuckooSearch::CuckooSearch()
{
    name_ = "CuckooSearch";
    selBestOp = static_cast<SelectionOperatorP> (new SelBestOp);
}
 
 
void CuckooSearch::registerParameters(StateP state)
{
    registerParameter(state, "pa", (voidP) new double(0.75), ECF::DOUBLE);
}
 
 
bool CuckooSearch::initialize(StateP state)
{
    selBestOp->initialize(state);
 
    voidP pDiscovery = getParameterValue(state, "pa");
    pa = *((double*)pDiscovery.get());
    if (pa < 0 || pa > 1)
    {
        ECF_LOG_ERROR(state, "Error - pa must be in interval [0,1]");
        throw "";
    }
 
    // reading boudaries and problem dimension
    voidP lBound = state->getGenotypes()[0]->getParameterValue(state, "lbound");
    lbound = *((double*)lBound.get());
    voidP uBound = state->getGenotypes()[0]->getParameterValue(state, "ubound");
    ubound = *((double*)uBound.get());
    voidP sptr = state->getGenotypes()[0]->getParameterValue(state, "dimension");
    numDimension = *((uint*)sptr.get());
 
    // algorithm accepts a single FloatingPoint or Binary genotype 
    // or a genotype derived from the abstract RealValueGenotype class
    GenotypeP activeGenotype = state->getGenotypes()[0];
    RealValueGenotypeP rv = boost::dynamic_pointer_cast<RealValueGenotype> (activeGenotype);
    if(!rv) {
        ECF_LOG_ERROR(state, "Error: Cuckoo Search algorithm accepts only a RealValueGenotype derived genotype! (FloatingPoint or Binary)");
        throw ("");
    }
 
    return true;
}
 
 
bool CuckooSearch::advanceGeneration(StateP state, DemeP deme)
{
    double sigma = 0.696574502;
    boost::mt19937 rng;
    boost::normal_distribution<> nd(0.0, 1.0);
    boost::variate_generator<boost::mt19937&,
    boost::normal_distribution<> > var_nor(rng, nd);
 
    IndividualP best = selBestOp->select(*deme);
    FloatingPointP bestFp = boost::static_pointer_cast<FloatingPoint::FloatingPoint> (best->getGenotype(0));
 
    // cuckoos via Levy flights (by Mantegna's algorithm)
    // new individual is added to population only if it is better than original individual
    for (uint i = 0; i < deme->size(); i++) {
        IndividualP trial = (IndividualP)deme->at(i)->copy();
        FloatingPointP trialFp = boost::static_pointer_cast<FloatingPoint::FloatingPoint> (trial->getGenotype(0));
        for (uint j = 0; j < numDimension; j++) {
            double u = var_nor() * sigma;
            double v = var_nor();
            double step = u / pow(fabs(v), 2 / (double)3);
            double randn = var_nor();
            double diff = trialFp->realValue[j] - bestFp->realValue[j];
            double stepsize = 0.01 * step * diff;
            trialFp->realValue[j] = trialFp->realValue[j] + stepsize*randn;
            if (trialFp->realValue[j] > ubound)
                trialFp->realValue[j] = ubound;
            if (trialFp->realValue[j] < lbound)
                trialFp->realValue[j] = lbound;
        }
        evaluate(trial);
        if (trial->fitness->isBetterThan(deme->at(i)->fitness))
            replaceWith(deme->at(i), trial);
    }
 
    // copy all individuals
    std::vector<IndividualP> nest1;
    std::vector<IndividualP> nest2;
    for (uint i = 0; i < deme->size(); i++) {
        IndividualP indCp = (IndividualP)deme->at(i)->copy();
        nest1.push_back(indCp);
        nest2.push_back(indCp);
    }
 
    // replace some individuals/nests by constructing new nests
    // nest is replaced only if it is better than original
    random_shuffle(nest1.begin(), nest1.end());
    random_shuffle(nest2.begin(), nest2.end());
    double randNum = (double)rand() / RAND_MAX;
    for (uint i = 0; i < deme->size(); i++) {
        IndividualP trial = (IndividualP)deme->at(i)->copy();
        FloatingPointP trialFp1 = boost::static_pointer_cast<FloatingPoint::FloatingPoint> (nest1.at(i)->getGenotype(0));
        FloatingPointP trialFp2 = boost::static_pointer_cast<FloatingPoint::FloatingPoint> (nest2.at(i)->getGenotype(0));
        FloatingPointP trialFp = boost::static_pointer_cast<FloatingPoint::FloatingPoint> (trial->getGenotype(0));
 
        for (uint j = 0; j < numDimension; j++) {
            if ((double)rand() / RAND_MAX < pa) {
                double stepsize = (trialFp1->realValue[j] - trialFp2->realValue[j])*randNum;
                trialFp->realValue[j] += stepsize;
                if (trialFp->realValue[j] > ubound)
                    trialFp->realValue[j] = ubound;
                if (trialFp->realValue[j] < lbound)
                    trialFp->realValue[j] = lbound;
            }
        }
 
        evaluate(trial);
        if (trial->fitness->isBetterThan(deme->at(i)->fitness))
            replaceWith(deme->at(i), trial);
    }
    return true;
 
}