/// Do the scaling on the population. Like the statistics and diversity, this
/// method does not change the contents of the population, but it does change
/// the values of the status members of the object. So we allow it to work on
/// a const population.
void
GAPopulation::scale(GABoolean flag) const
{
if(scaled == gaTrue && flag != gaTrue)
{
return;
}
GAPopulation* This = (GAPopulation*)this;
if(n > 0)
{
sclscm->evaluate(*This);
float tmpsum;
This->fitMin = This->fitMax = tmpsum = sind[0]->fitness();
unsigned int i;
for(i = 1; i < n; i++)
{
tmpsum += sind[i]->fitness();
This->fitMax = GAMax(fitMax, sind[i]->fitness());
This->fitMin = GAMin(fitMin, sind[i]->fitness());
}
float tmpave = tmpsum / n;
This->fitAve = tmpave;
This->fitSum = tmpsum; /// if scores are huge we'll lose data here
float tmpvar = 0.0;
if(n > 1)
{
for(i = 0; i < n; i++)
{
float s = sind[i]->fitness() - This->fitAve;
s *= s;
tmpvar += s;
}
tmpvar /= (n - 1);
}
This->fitDev = (float)sqrt(tmpvar);
This->fitVar = tmpvar; /// could lose data if huge variance
}
else
{
This->fitMin = This->fitMax = This->fitSum = 0.0;
This->fitVar = This->fitDev = 0.0;
}
This->scaled = gaTrue;
This->ssorted = gaFalse;
}
/// Evaluate each member of the population and store basic population statistics
/// in the member variables. It is OK to run this on a const object - it
/// changes to physical state of the population, but not the logical state.
/// The partial sums are normalized to the range [0,1] so that they can be
/// used whether the population is sorted as low-is-best or high-is-best.
/// Individual 0 is always the best individual, and the partial sums are
/// calculated so that the worst individual has the smallest partial sum. All
/// of the partial sums add to 1.0.
void
GAPopulation::statistics(GABoolean flag) const
{
if(statted == gaTrue && flag != gaTrue)
{
return;
}
GAPopulation * This = (GAPopulation *)this;
if(n > 0)
{
float tmpsum;
This->rawMin = This->rawMax = tmpsum = rind[0]->score();
unsigned int i;
for(i = 1; i < n; i++)
{
float scr = rind[i]->score();
tmpsum += scr;
This->rawMax = GAMax(rawMax, scr);
This->rawMin = GAMin(rawMin, scr);
}
float tmpave = tmpsum / n;
This->rawAve = tmpave;
This->rawSum = tmpsum; /// if scores are huge we'll lose data here
float tmpvar = 0.0;
if(n > 1)
{
for(i = 0; i < n; i++)
{
float s = rind[i]->score() - This->rawAve;
s *= s;
tmpvar += s;
}
tmpvar /= (n - 1);
}
This->rawDev = (float)sqrt(tmpvar);
This->rawVar = tmpvar; /// could lose data if huge variance
}
else
{
This->rawMin = This->rawMax = This->rawSum = 0.0;
This->rawDev = This->rawVar = 0.0;
}
This->statted = gaTrue;
}
// Make sure our bitmask is big enough, generate a mask, then use it to
// extract the information from each parent to stuff the two children.
// We don't deallocate any space for the masks under the assumption that we'll
// have to use them again in the future.
// For now we'll implement this only for fixed length genomes. If you use
// this crossover method on genomes of different sizes it might break!
int
GA3DBinaryStringGenome::
UniformCrossover(const GAGenome& p1, const GAGenome& p2,
GAGenome* c1, GAGenome* c2){
GA3DBinaryStringGenome &mom=(GA3DBinaryStringGenome &)p1;
GA3DBinaryStringGenome &dad=(GA3DBinaryStringGenome &)p2;
int i,j,k, nc=0;;
if(c1 && c2){
GA3DBinaryStringGenome &sis=(GA3DBinaryStringGenome &)*c1;
GA3DBinaryStringGenome &bro=(GA3DBinaryStringGenome &)*c2;
if(sis.width() == bro.width() && sis.height() == bro.height() &&
sis.depth() == bro.depth() &&
mom.width() == dad.width() && mom.height() == dad.height() &&
mom.depth() == dad.depth() &&
sis.width() == mom.width() && sis.height() == mom.height() &&
sis.depth() == mom.depth()){
for(i=sis.width()-1; i>=0; i--){
for(j=sis.height()-1; j>=0; j--){
for(k=sis.depth()-1; k>=0; k--){
if(GARandomBit()){
sis.gene(i,j,k, mom.gene(i,j,k));
bro.gene(i,j,k, dad.gene(i,j,k));
}
else{
sis.gene(i,j,k, dad.gene(i,j,k));
bro.gene(i,j,k, mom.gene(i,j,k));
}
}
}
}
}
else{
GAMask mask;
int maxx = GAMax(sis.width(), bro.width());
int minx = GAMin(mom.width(), dad.width());
int maxy = GAMax(sis.height(), bro.height());
int miny = GAMin(mom.height(), dad.height());
int maxz = GAMax(sis.depth(), bro.depth());
int minz = GAMin(mom.depth(), dad.depth());
mask.size(maxx*maxy*maxz);
for(i=0; i<maxx; i++)
for(j=0; j<maxy; j++)
for(k=0; k<maxz; k++)
mask[i*maxy*maxz+j*maxz+k] = GARandomBit();
minx = GAMin(sis.width(), minx);
miny = GAMin(sis.height(), miny);
minz = GAMin(sis.depth(), minz);
for(i=minx-1; i>=0; i--)
for(j=miny-1; j>=0; j--)
for(k=minz-1; k>=0; k--)
sis.gene(i,j,k, (mask[i*miny*minz+j*minz+k] ?
mom.gene(i,j,k) : dad.gene(i,j,k)));
minx = GAMin(bro.width(), minx);
miny = GAMin(bro.height(), miny);
minz = GAMin(bro.depth(), minz);
for(i=minx-1; i>=0; i--)
for(j=miny-1; j>=0; j--)
for(k=minz-1; k>=0; k--)
bro.gene(i,j,k, (mask[i*miny*minz+j*minz+k] ?
dad.gene(i,j,k) : mom.gene(i,j,k)));
}
nc = 2;
}
else if(c1 || c2){
GA3DBinaryStringGenome &sis =
(c1 ? (GA3DBinaryStringGenome&)*c1 : (GA3DBinaryStringGenome&)*c2);
if(mom.width() == dad.width() && mom.height() == dad.height() &&
mom.depth() == dad.depth() &&
sis.width() == mom.width() && sis.height() == mom.height() &&
sis.depth() == mom.depth()){
for(i=sis.width()-1; i>=0; i--)
for(j=sis.height()-1; j>=0; j--)
for(k=sis.depth()-1; k>=0; k--)
sis.gene(i,j,k, (GARandomBit() ? mom.gene(i,j,k):dad.gene(i,j,k)));
}
else{
int minx = GAMin(mom.width(), dad.width());
int miny = GAMin(mom.height(), dad.height());
int minz = GAMin(mom.depth(), dad.depth());
minx = GAMin(sis.width(), minx);
miny = GAMin(sis.height(), miny);
minz = GAMin(sis.depth(), minz);
for(i=minx-1; i>=0; i--)
for(j=miny-1; j>=0; j--)
for(k=minz-1; k>=0; k--)
sis.gene(i,j,k, (GARandomBit() ? mom.gene(i,j,k):dad.gene(i,j,k)));
}
nc = 1;
}
return nc;
}
