population.cpp

Go to the documentation of this file.

#include "population.h"
 
namespace Brush{   
namespace Pop{
        
template<ProgramType T>
Population<T>::Population()
{
    individuals.resize(0);
    mig_prob = 0.0;
    pop_size = 0;
    num_islands = 0;
    linear_complexity = false;
}
 
 
template<ProgramType T>
void Population<T>::init(vector<Individual<T>>& new_individuals, const Parameters& params)
{
    if (new_individuals.size() != params.pop_size
    &&  new_individuals.size() != 2*params.pop_size ) {
        throw std::runtime_error("Individual vector has different number of individuals than pop_size. popsize is "+to_string(params.pop_size)+", number of individuals is " + to_string(new_individuals.size()));
    }
 
    this->mig_prob = params.mig_prob;
    this->pop_size = params.pop_size;
    this->num_islands=params.num_islands;
    this->linear_complexity = in(params.objectives, std::string("linear_complexity"));
 
    island_indexes.resize(num_islands);
    
    // If the assert fails, execution stops, but for completeness, you can also throw an exception
    size_t p = pop_size;
 
    individuals.resize(2*p);
    std::fill(individuals.begin(), individuals.end(), nullptr);
 
    for (int i=0; i<num_islands; ++i)
    {
        size_t idx_start = std::floor(i*p/num_islands);
        size_t idx_end   = std::floor((i+1)*p/num_islands);
 
        auto delta = idx_end - idx_start;
 
        island_indexes.at(i).resize(delta);
        iota(island_indexes.at(i).begin(), island_indexes.at(i).end(), idx_start);
    
        if (new_individuals.size() == 2*params.pop_size) { // pop + offspring
            island_indexes.at(i).resize(delta*2);
            iota(
                island_indexes.at(i).begin() + delta, island_indexes.at(i).end(),
                p+idx_start);
        }
    };
 
    for (int j=0; j< new_individuals.size(); j++) {
        individuals.at(j) = std::make_shared<Individual<T>>(new_individuals.at(j));
    }
}
 
template<ProgramType T>
void Population<T>::init(SearchSpace& ss, const Parameters& params)
{
    this->mig_prob = params.mig_prob;
    this->pop_size = params.pop_size;
    this->num_islands=params.num_islands;
    this->linear_complexity = in(params.objectives, std::string("linear_complexity"));
    
    // Tuples with start and end indexes for each island. Number of individuals
    // in each island can slightly differ if num_islands is not a divisor of p (popsize)
    island_indexes.resize(num_islands);
    
    size_t p = pop_size; // population size
 
    for (int i=0; i<num_islands; ++i)
    {
        size_t idx_start = std::floor(i*p/num_islands);
        size_t idx_end   = std::floor((i+1)*p/num_islands);
        
        auto delta = idx_end - idx_start;
 
        island_indexes.at(i).resize(delta);
        iota(island_indexes.at(i).begin(), island_indexes.at(i).end(), idx_start);
    };
 
    // this calls the default constructor for the container template class 
    individuals.resize(2*p); // we will never increase or decrease the size during execution (because is not thread safe). this way, theres no need to sync between selecting and varying the population
 
    for (int i = 0; i< p; ++i)
    {          
        // first half will contain the initial population
        individuals.at(i) = std::make_shared<Individual<T>>();
        individuals.at(i)->init(ss, params);
        
        // second half is space to the offspring (but we dont initialize them)
        individuals.at(p+i) = nullptr;
    }
}
 
template<ProgramType T>
void Population<T>::save(string filename)
{
    std::ofstream out;                      
    if (!filename.empty())
        out.open(filename);
    else
        out.open("population.json");
 
    json j;
    to_json(j, *this);
    out << j ;
    out.close();
    logger.log("Saved population to file " + filename, 1);
}
 
template<ProgramType T>
void Population<T>::load(string filename)
{
    std::ifstream indata;
    indata.open(filename);
    if (!indata.good())
        HANDLE_ERROR_THROW("Invalid input file " + filename + "\n"); 
 
    std::string line;
    indata >> line; 
 
    json j = json::parse(line);
    from_json(j, *this);
 
    logger.log("Loaded population from " + filename + " of size = " 
            + to_string(this->size()),1);
 
    indata.close();
}

template<ProgramType T>
void Population<T>::add_offspring_indexes(int island)
{      
    size_t p = pop_size; // population size. prep_offspring slots will douple the population, adding the new expressions into the islands
    
    // this is going to be tricky (pay attention to delta and p use)
    size_t idx_start = std::floor(island*p/num_islands);
    size_t idx_end   = std::floor((island+1)*p/num_islands);
 
    auto delta = idx_end - idx_start; // island size
 
    // inserting indexes of the offspring
    island_indexes.at(island).resize(island_indexes.at(island).size() + delta);
    iota(
        island_indexes.at(island).begin() + delta, island_indexes.at(island).end(),
        p+idx_start);
 
    // Im keeping the offspring and parents in the same population object, because we
    // have operations that require them together (archive, hall of fame.)
    // The downside is having to be aware that islands will create offsprings
    // intercalated with other islands
}
 
template<ProgramType T>
void Population<T>::update(vector<vector<size_t>> survivors)
{
 
    // this is the step that should end up cutting off half of the population
    vector<Individual<T>> new_pop;
    new_pop.resize(0);
    for (int k=0; k<survivors.at(0).size(); ++k){
        new_pop.push_back( *individuals.at(survivors.at(0).at(k)) );
    }
 
    assert(new_pop.size() == individuals.size() && " migration ended up with a different popsize");
 
    for (int j=0; j<num_islands; ++j)
    {
 
        // need to make island point to original range
        size_t idx_start = std::floor(j*pop_size/num_islands);
        size_t idx_end   = std::floor((j+1)*pop_size/num_islands);
        auto delta = idx_end - idx_start;
 
        // inserting indexes of the offspring
        island_indexes.at(j).clear();
        island_indexes.at(j).resize(delta);
        iota(island_indexes.at(j).begin(), island_indexes.at(j).end(), idx_start);
    }
 
    assert(new_pop.size() == pop_size
           && " update ended up with a different popsize");
 
    this->individuals.resize(0);
    for (auto ind : new_pop)
    {
        // making hard copies of the individuals
        json ind_copy = ind;
 
        // this will fill just half of the individuals vector
        individuals.push_back(
            std::make_shared<Individual<T>>(ind_copy) );
    }
 
    assert(individuals.size() == pop_size
           && " number of new individuals is different from pop size");
 
    for (int i=0; i< pop_size; ++i)
    {
        // second half is space to the offspring (but we dont initialize them)
        individuals.push_back(nullptr);   
    }
}
 
template<ProgramType T>
string Population<T>::print_models(string sep)
{
    // TODO: rename it. This function does not print anything, just returns a string
    string output = "";
 
    for (int j=0; j<num_islands; ++j)
    {
        for (int k=0; k<island_indexes.at(j).size(); ++k) {
            Individual<T>& ind = *individuals.at(island_indexes.at(j).at(k)).get();
 
            output += "island " + to_string(j) + " ";
            output += "index " + to_string(island_indexes.at(j).at(k)) + ": ";
            output += ind.get_model() + " ";
            output += ind.fitness.toString() + sep;
        }
    }
    return output;
}
 
template<ProgramType T>
vector<vector<size_t>> Population<T>::sorted_front(unsigned rank)
{
    // this is used to migration and update archive at the end of a generation. expect islands without offspring
 
    /* Returns individuals on the Pareto front, sorted by increasign complexity. */
    vector<vector<size_t>> pf_islands;
    pf_islands.resize(num_islands);
 
    for (int j=0;j<num_islands; ++j)
    {
        auto indices = island_indexes.at(j);
        vector<size_t> pf;
 
        for (int i=0; i<indices.size(); ++i)
        {
            // this assumes that rank was previously calculated. It is set in selection (ie nsga2) if the information is useful to select/survive
            if (individuals.at(indices.at(i))->fitness.rank == rank)
                pf.push_back(i);
        }
 
        if (this->linear_complexity)
            std::sort(pf.begin(),pf.end(),SortLinearComplexity(*this)); 
        else
            std::sort(pf.begin(),pf.end(),SortComplexity(*this)); 
 
        auto it = std::unique(pf.begin(),pf.end(),SameFitComplexity(*this));
        
        pf.resize(std::distance(pf.begin(),it));
        pf_islands.at(j) = pf;
    }
 
    return pf_islands;
}
 
template<ProgramType T>
vector<size_t> Population<T>::hall_of_fame(unsigned rank)
{
    // Inspired in fast nds from nsga2
 
    // TODO: use hall of fame instead of re-implmementing this feature in
    // archive init and update functions
 
    // this is used to migration and update archive at the end of a generation.
    // Thiis function expects islands without offspring
 
    // TODO: run fast_nds here to get the pareto fronts? I think this could improve performance
    vector<size_t> merged_islands(0);
    
    for (int j=0;j<num_islands; ++j)
    {
        auto indices = island_indexes.at(j);
        for (int i=0; i<indices.size(); ++i)
        {
            merged_islands.push_back(indices.at(i));
        }
    }
 
    // checking if there is no dominance between different fronts
    // (without updating their fitness objects)
    vector<size_t> hof;                
    hof.clear();
 
    for (int i = 0; i < merged_islands.size(); ++i) {
        int dcount = 0;
    
        auto p = individuals.at(merged_islands[i]);
 
        for (int j = 0; j < merged_islands.size(); ++j) {
            const Individual<T>& q = (*individuals.at(merged_islands[j]));
        
            int compare = p->fitness.dominates(q.fitness);
            if (compare == -1) { // q dominates p
                //p.dcounter += 1;
                dcount += 1;
            }
        }
 
        if (dcount == 0) {
            hof.push_back(merged_islands[i]);
        }
    }
 
    if (this->linear_complexity)
        std::sort(hof.begin(),hof.end(),SortLinearComplexity(*this)); 
    else
        std::sort(hof.begin(),hof.end(),SortComplexity(*this)); 
        
    auto it = std::unique(hof.begin(),hof.end(),SameFitComplexity(*this));
    
    hof.resize(std::distance(hof.begin(),it));
 
    return hof;
}
 
template<ProgramType T>
void Population<T>::migrate()
{
    // changes where island points to by shuffling it
 
    if (num_islands==1)
        return; // skipping. this only work because update is fixing island indexes
 
    // This method is not thread safe (as it is now)
    vector<vector<size_t>> new_island_indexes;
    new_island_indexes.resize(num_islands);
 
    for (int island=0; island<num_islands; ++island)
    {
        new_island_indexes.at(island).resize(0);
 
        auto indices = island_indexes.at(island);
        for (unsigned int i=0; i<indices.size(); ++i)
        {
            if (r() < mig_prob)
            {
                size_t migrating_idx;
                
                vector<int> other_islands(num_islands-1);
                iota(other_islands.begin(), other_islands.end(), 0);
 
                // skipping current island
                auto it = other_islands.begin();
                std::advance(it, island);
                for (;it != other_islands.end(); ++it) {
                    ++(*it);
                }
 
                // picking other island
                int other_island = *r.select_randomly(
                    other_islands.begin(),
                    other_islands.end());
 
                migrating_idx = *r.select_randomly(
                    island_indexes.at(other_island).begin(),
                    island_indexes.at(other_island).end());
 
                new_island_indexes.at(island).push_back(migrating_idx);
            }
            else
            {
                new_island_indexes.at(island).push_back(indices.at(i));
            }
        }
    }
 
    // making hard copies (so the next generation starts with islands that does not share individuals 
    // this is particularly important to avoid multiple threads assigning different rank/crowdist/dcounter 
    // or different fitness)
 
    vector<Individual<T>> new_pop;
    new_pop.resize(0);
    for (int j=0; j<num_islands; ++j)
    {
        for (int k=0; k<new_island_indexes.at(j).size(); ++k){
            new_pop.push_back(
                *individuals.at(new_island_indexes.at(j).at(k)) );
        }
 
        // need to make island point to original range
        size_t idx_start = std::floor(j*pop_size/num_islands);
        size_t idx_end   = std::floor((j+1)*pop_size/num_islands);
 
        auto delta = idx_end - idx_start;
 
        assert(delta == new_island_indexes.at(j).size()
            && " new pop has the wrong number of new individuals");
 
        // inserting indexes of the offspring
        island_indexes.at(j).clear();
        island_indexes.at(j).resize(delta);
        iota(island_indexes.at(j).begin(), island_indexes.at(j).end(), idx_start);
    }
 
    assert(new_pop.size() == pop_size
           && " migration ended up with a different popsize");
 
    this->individuals.resize(0);
    for (auto ind : new_pop)
    {
        // making hard copies of the individuals
        json ind_copy = ind;
 
        // this will fill just half of the pop
        individuals.push_back(
            std::make_shared<Individual<T>>(ind_copy) );
    }
    for (int i=0; i< pop_size; ++i)
    {
        // second half is space to the offspring (but we dont initialize them)
        individuals.push_back(nullptr);   
    }
}
 
} // Pop
} // Brush