void GeneticAlgorithm::createOffspring(Population& newPopulation,
bool doMutation) {
Image& targetImage = *TargetImage::Instance();
Population tmpResult;
while (m_population.size() > 1) {
m_inputLock.lock();
if (m_population.size() <= 1) {
m_inputLock.unlock();
break;
}
std::pair<Organism*, Organism*> couple =
m_pairGenerator.removeRandomPair(m_population);
m_inputLock.unlock();
Organism* child = new Organism(*couple.first, *couple.second, doMutation);
double score = averagePixelDistance(targetImage, child->getPhenotype());
child->setScore(score);
tmpResult.push_back(child);
tmpResult.push_back(couple.first);
tmpResult.push_back(couple.second);
}
m_outputLock.lock();
newPopulation.insert(newPopulation.end(), tmpResult.begin(),
tmpResult.end());
m_outputLock.unlock();
}
void Crossover::do_crossover(Population& population) {
shuffle(population.begin(), population.end(), rng);
for (int index = 0; index + 1 < int(population.size()); index += 2)
if (decide_to_do_crossover()) {
auto crossed = do_crossover(population[index], population[index + 1]);
show_crossing(crossed);
}
}
void Merge(const Population& population) {
samples.resize(children_start + population.size());
copy(population.begin(), population.end(), samples.begin()+children_start);
sort(samples.begin(), samples.end());
samples.resize(capacity());
children_start = capacity();
}
void PerPointStopCriterion<TFunction>::update(const Population& population) {
for (Population::Iterator it = population.begin();
it != population.end();
++it) {
typename OnePointStopCriterion<TFunction>::Ptr one_point_stop_criterion =
stop_criterion_info_->getInfoById(it.point_id());
one_point_stop_criterion->update(it.point());
}
}
void
APG::ConstraintSolver::run()
{
if ( !m_readyToRun ) {
error() << "DANGER WILL ROBINSON! A ConstraintSolver (serial no:" << m_serialNumber << ") tried to run before its QueryMaker finished!";
m_abortRequested = true;
return;
}
if ( m_domain.empty() ) {
debug() << "The QueryMaker returned no tracks";
return;
} else {
debug() << "Domain has" << m_domain.size() << "tracks";
}
debug() << "Running ConstraintSolver" << m_serialNumber;
emit totalSteps( m_maxGenerations );
// GENETIC ALGORITHM LOOP
Population population;
quint32 generation = 0;
Meta::TrackList* best = NULL;
while ( !m_abortRequested && ( generation < m_maxGenerations ) ) {
quint32 s = m_constraintTreeRoot->suggestPlaylistSize();
m_suggestedPlaylistSize = (s > 0) ? s : m_suggestedPlaylistSize;
fill_population( population );
best = find_best( population );
if ( population.value( best ) < m_satisfactionThreshold ) {
select_population( population, best );
mutate_population( population );
generation++;
emit incrementProgress();
} else {
break;
}
}
debug() << "solution at" << (void*)(best);
m_solvedPlaylist = best->mid( 0 );
m_finalSatisfaction = m_constraintTreeRoot->satisfaction( m_solvedPlaylist );
/* clean up */
Population::iterator it = population.begin();
while ( it != population.end() ) {
delete it.key();
it = population.erase( it );
}
emit endProgressOperation( this );
}
Population run_island(Configuration conf, Individual initial_individual){
// Random Number Generation
std::default_random_engine eng(std::random_device{}());
std::uniform_real_distribution<> dist(-5, 5);
// population generation
Population population;
population.push_back(initial_individual); // push the initial as it is.
for (int i=1; i<conf.pop_size; i++) {
Individual indi = mutate(conf, initial_individual);
population.push_back(indi);
}
int generations = conf.num_generations;
for (int g=0; g<generations; g++) {
// first fitness calculation
for (int i=0; i<conf.pop_size; i++) {
population[i].score = fitness(population[i]);
}
// do sorting according to score
std::sort(population.begin(), population.end(), compare);
if (g != (conf.num_generations - 1)) {
// crossover
for (int i=0; i<(conf.pop_size / 2); i++) {
crossover(conf, population[2*i], population[2*i+1]);
}
// mutate
for (int i=0; i<conf.pop_size; i++) {
population[i] = mutate(conf, population[i]);
}
}
}
return population;
}
/*
* Use the genetic algorithm to construct a new population from the old
*/
Population GeneticPool::epoch(Population& old_pop, const BattleFieldLayout& bfl) {
if (!initialized_ || old_pop.size() == 1) {
Population new_pop = old_pop;
new_pop.clear();
old_pop.stats_.reset();
for (Ship& t : old_pop) {
t.calculateFitness(bfl, old_pop.facilities_.front(), old_pop.winner_);
t.isElite = false;
}
//sort the population (for scaling and elitism)
sort(old_pop.begin(), old_pop.end());
//calculate best, worst, average and total fitness
calculateStatistics(old_pop);
for (Ship& t : old_pop) {
new_pop.push_back(t.makeChild());
}
new_pop.stats_ = old_pop.stats_;
return new_pop;
}
Population new_pop = old_pop;
new_pop.clear();
old_pop.stats_.reset();
for (Ship& t : old_pop) {
t.calculateFitness(bfl, old_pop.facilities_.front(), old_pop.winner_);
t.isElite = false;
}
//sort the population (for scaling and elitism)
sort(old_pop.begin(), old_pop.end());
//calculate best, worst, average and total fitness
calculateStatistics(old_pop);
/*
* Only winning teams get to have an elite by copying
* some of the fittest genomes without any mutation/crossover.
* Make sure we add an EVEN number or the roulette wheel sampling will crash
* NOTE: we don't dont copy elites if we use perf descriptors since that is already a form of elitism
*/
if (old_pop.winner_ && !layout_.usePerfDesc_ && layout_.numElite_ < old_pop.size()) {
if (!(layout_.numEliteCopies_ * (layout_.numElite_ % 2))) {
copyNBest(layout_.numElite_, layout_.numEliteCopies_, old_pop, new_pop);
}
}
CHECK(old_pop.layout_.sl_.numPerfDesc_ == 4);
PerfDescBsp<4, Ship> pdb;
if (layout_.usePerfDesc_) {
for (Ship& t : old_pop) {
CHECK(t.perfDesc_.size() == 4);
pdb.insert(&t);
}
}
//now we enter the GA loop
//repeat until a new population is generated
while (new_pop.size() < old_pop.size()) {
//grab two chromosones
Ship& mum = pickSpecimen(old_pop);
Ship* dad;
if (layout_.usePerfDesc_) {
CHECK(old_pop.size() > 2);
auto pair = pdb.findClosestMate(&mum);
Ship* closest = *pair.first;
Coord dist = pair.second;
if (dist == 0.0)
dist = 0.01;
vector<Ship*> result;
Coord range = dist * 1.5;
do {
result = pdb.findInRange(&mum, range);
range *= 1.5;
} while (result.size() < 2);
/*for(Ship* s : result) {
std::cerr << s->perfDesc_[0] << '\t' << s->perfDesc_[2] << '\t' << s->perfDesc_[3] << '\t' << s->perfDesc_[4] << std::endl;
}*/
if (result.empty()) {
dad = closest;
} else {
//std::cerr << "dist/range:" << dist << "\t" << range << std::endl;
std::sort(result.begin(), result.end(), [&](const Ship* s1, const Ship* s2) {
return s1->fitness_ < s2->fitness_;
});
dad = pickSpecimen(result);
}
} else
dad = &pickSpecimen(old_pop);
//create some offspring via crossover
std::pair<Ship, Ship> babies = crossover(mum, *dad, layout_.crossoverIterations);
//now we mutate
mutate(*babies.first.brain_);
mutate(*babies.second.brain_);
//now copy into vecNewPop population
new_pop.push_back(babies.first);
if (new_pop.size() < old_pop.size())
new_pop.push_back(babies.second);
}
CHECK(new_pop.size() == old_pop.size());
old_pop.stats_.generationCnt_++;
new_pop.stats_ = old_pop.stats_;
return new_pop;
}
