FloatingPointCrsBga.cpp

#include "../ECF_base.h"
#include "FloatingPoint.h"
#include <math.h>
 
 
namespace FloatingPoint
{
 
void FloatingPointCrsBga::registerParameters(StateP state)
{
    myGenotype_->registerParameter(state, "crx.bga", (voidP) new double(0), ECF::DOUBLE);
}
 
 
bool FloatingPointCrsBga::initialize(StateP state)
{
    voidP sptr = myGenotype_->getParameterValue(state, "crx.bga");
    probability_ = *((double*)sptr.get());
 
    return true;
}
 
 
bool FloatingPointCrsBga::mate(GenotypeP gen1, GenotypeP gen2, GenotypeP child)
{
    FloatingPoint* p1 = (FloatingPoint*) (gen1.get());
    FloatingPoint* p2 = (FloatingPoint*) (gen2.get());
    FloatingPoint* ch = (FloatingPoint*) (child.get());
 
    int a;
    uint size = (uint) p1->realValue.size();
    // upper - lower bound
    double range = 0.5 * (p1->getUBound() - p1->getLBound());
    double gama = 0, lambda = 0, b;
 
    for (int i = 0; i <= 15; i++) {
        a = state_->getRandomizer()->getRandomInteger(1, 16);
        // 1/16 probability for 1 = 1
        if (a == 16)
            a = 1;
        else
            a = 0;
        gama = gama + a * pow((double) 2., -i);
    }
 
    double norm = 0;
    for(uint i = 0; i < size; i++)
        norm += pow(p1->realValue[i] - p2->realValue[i], 2);
    norm = sqrt(norm);
 
    // scaling safeguard 
    if(norm < 10e-9)
        norm = 1;
 
    // determine better parent
    FloatingPoint *better, *worse;
    FitnessP parent2 = state_->getContext()->secondParent->fitness;
    if(state_->getContext()->firstParent->fitness->isBetterThan(parent2)) {
        better = p1;
        worse = p2;
    } else {
        better = p2;
        worse = p1;
    }
 
    // build child
    for (uint i = 0; i < size; i++) {
        // worse gene minus better gene divided with norm
        lambda = (worse->realValue[i] - better->realValue[i]) / norm;
 
        b = state_->getRandomizer()->getRandomDouble();
        // minus with probability 0.9
        if (b <= 0.9)
            ch->realValue[i] = better->realValue[i] - range * gama * lambda;
        else
            ch->realValue[i] = better->realValue[i] + range * gama * lambda;
 
        // check for bounds; if outside the interval, bring closer to better parent
        if(ch->realValue[i] > ch->getUBound())
            ch->realValue[i] = better->realValue[i] + state_->getRandomizer()->getRandomDouble() * (ch->getUBound() - better->realValue[i]);
        else if(ch->realValue[i] < ch->getLBound())
            ch->realValue[i] = better->realValue[i] - state_->getRandomizer()->getRandomDouble() * (better->realValue[i] - ch->getLBound());
    }
 
    return true;
}
 
}