population random_breeder::breed(const population &generation, const std::vector<double> &fitnesses) const
{
using namespace std;
population ret;
tree_crossover_reproducer repr;
node_crossover_reproducer mrepr;
builder random_builder;
experimental_parameters e = exp_params();
const uint32_t growth_tree_min = e["growth_tree_min"];
const uint32_t growth_tree_max = e["growth_tree_max"];
while(ret.size() < generation.size())
{
const agent &mother = generation[rand() % generation.size()];
const agent &father = generation[rand() % generation.size()];
agent mutant = random_builder.grow(program_types_set, growth_tree_min, growth_tree_max);
const vector<agent> children = repr.reproduce(vector<agent> { mother, father });
if(children.size() > 0)
{
agent res0 = mrepr.reproduce(vector<agent> { children[0], mutant })[0];
ret.push_back(res0);
}
if(children.size() > 1)
{
agent res1 = mrepr.reproduce(vector<agent> { children[1], mutant })[0];
ret.push_back(res1);
}
}
return ret;
}
family_vector rnd_selector::select(const population&p,const int fcount)
{
family_vector result(fcount);
for(int i=0;i<fcount;i++)
result[i]=std::make_pair(m_random.uniform(0,p.size()-1),
m_random.uniform(0,p.size()-1));
return result;
}
void adapt_population(population& p)
{
float F_min = min_element(p.begin(),p.end(),eval_comp)->eval;
for(unsigned int i=0; i<p.size(); i++)
{
float sum = 0.0;
for(unsigned int j=0; j<p.size(); j++)
sum += p[j].eval - F_min;
p[i].adapt = (p[i].eval - F_min)/sum;
}
}
void report(population& p)
{
switch(cfg.report_every)
{
case config::report::none:
break;
case config::report::avg:
{
float avg = 0.0;
for(auto i = p.begin(); i != p.end(); ++i)
avg += i->eval;
avg /= static_cast<float>(p.size());
std::cout << iter << ' ' << avg << '\n';
}
break;
case config::report::best:
std::cout << iter << ' ' << p[0].eval << '\n';
break;
}
if(cfg.report_var)
{
std::cout << iter << ' ' << statistics::variance(p) << '\n';
}
}
std::vector<std::pair<population::size_type,std::vector<population::individual_type>::size_type> >
random_r_policy::select(const std::vector<population::individual_type> &immigrants, const population &dest) const
{
const population::size_type rate_limit = std::min<population::size_type>(get_n_individuals(dest),boost::numeric_cast<population::size_type>(immigrants.size()));
// Temporary vectors to store sorted indices of the populations.
std::vector<population::size_type> immigrants_idx(boost::numeric_cast<std::vector<population::size_type>::size_type>(immigrants.size()));
std::vector<population::size_type> dest_idx(boost::numeric_cast<std::vector<population::size_type>::size_type>(dest.size()));
// Fill in the arrays of indices.
iota(immigrants_idx.begin(),immigrants_idx.end(),population::size_type(0));
iota(dest_idx.begin(),dest_idx.end(),population::size_type(0));
// Permute the indices (immigrants).
for (population::size_type i = 0; i < rate_limit; ++i) {
population::size_type next_idx = i + (m_urng() % (rate_limit - i));
if (next_idx != i) {
std::swap(immigrants_idx[i], immigrants_idx[next_idx]);
}
}
// Permute the indices (destination).
for (population::size_type i = 0; i < rate_limit; ++i) {
population::size_type next_idx = i + (m_urng() % (dest.size() - i));
if (next_idx != i) {
std::swap(dest_idx[i], dest_idx[next_idx]);
}
}
// Return value.
std::vector<std::pair<population::size_type,std::vector<population::individual_type>::size_type> >
retval;
for (population::size_type i = 0; i < rate_limit; ++i) {
retval.push_back(std::make_pair(dest_idx[i],immigrants_idx[i]));
}
return retval;
}
/**
* @param[in] pop input population.
*
* @return the number of individuals to be migrated from/to the population according to the current migration rate and type.
*/
population::size_type base::get_n_individuals(const population &pop) const
{
population::size_type retval = 0;
switch (m_type) {
case absolute:
retval = boost::numeric_cast<population::size_type>(m_rate);
if (retval > pop.size()) {
pagmo_throw(value_error,"absolute migration rate is higher than population size");
}
break;
case fractional:
retval = boost::numeric_cast<population::size_type>(double_to_int::convert(m_rate * pop.size()));
pagmo_assert(retval <= pop.size());
}
return retval;
}
void ms::evolve(population &pop) const
{
// Let's store some useful variables.
const population::size_type NP = pop.size();
// Get out if there is nothing to do.
if (m_starts == 0 || NP == 0) {
return;
}
// Local population used in the algorithm iterations.
population working_pop(pop);
//ms main loop
for (int i=0; i< m_starts; ++i)
{
working_pop.reinit();
m_algorithm->evolve(working_pop);
if (working_pop.problem().compare_fc(working_pop.get_individual(working_pop.get_best_idx()).cur_f,working_pop.get_individual(working_pop.get_best_idx()).cur_c,
pop.get_individual(pop.get_worst_idx()).cur_f,pop.get_individual(pop.get_worst_idx()).cur_c
) )
{
//update best population replacing its worst individual with the good one just produced.
pop.set_x(pop.get_worst_idx(),working_pop.get_individual(working_pop.get_best_idx()).cur_x);
pop.set_v(pop.get_worst_idx(),working_pop.get_individual(working_pop.get_best_idx()).cur_v);
}
if (m_screen_output)
{
std::cout << i << ". " << "\tCurrent iteration best: " << working_pop.get_individual(working_pop.get_best_idx()).cur_f << "\tOverall champion: " << pop.champion().f << std::endl;
}
}
}
/// Evolve method.
void monte_carlo::evolve(population &pop) const
{
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const problem::base::size_type prob_dimension = prob.get_dimension(), prob_i_dimension = prob.get_i_dimension();
const decision_vector &lb = prob.get_lb(), &ub = prob.get_ub();
const population::size_type pop_size = pop.size();
// Get out if there is nothing to do.
if (pop_size == 0 || m_max_eval == 0) {
return;
}
// Initialise temporary decision vector, fitness vector and decision vector.
decision_vector tmp_x(prob_dimension);
fitness_vector tmp_f(prob.get_f_dimension());
constraint_vector tmp_c(prob.get_c_dimension());
// Main loop.
for (std::size_t i = 0; i < m_max_eval; ++i) {
// Generate a random decision vector.
for (problem::base::size_type j = 0; j < prob_dimension - prob_i_dimension; ++j) {
tmp_x[j] = boost::uniform_real<double>(lb[j],ub[j])(m_drng);
}
for (problem::base::size_type j = prob_dimension - prob_i_dimension; j < prob_dimension; ++j) {
tmp_x[j] = boost::uniform_int<int>(lb[j],ub[j])(m_urng);
}
// Compute fitness and constraints.
prob.objfun(tmp_f,tmp_x);
prob.compute_constraints(tmp_c,tmp_x);
// Locate the worst individual.
const population::size_type worst_idx = pop.get_worst_idx();
if (prob.compare_fc(tmp_f,tmp_c,pop.get_individual(worst_idx).cur_f,pop.get_individual(worst_idx).cur_c)) {
pop.set_x(worst_idx,tmp_x);
}
}
}
void reallocate_stra(population &pop,int stra_idx,int size,double val)
{
int pop_size=pop.size();
int i;
for ( i=0;i<pop_size;i++ )
pop[i].stra[stra_idx].assign(size,val);
}
void allocate_pop(population &pop,size_t NumDims,int stra_num,int num_obj)
{
size_t pop_size=pop.size();
size_t i;
for ( i=0;i<pop_size;i++ )
allocate_ind(pop[i],NumDims,stra_num,num_obj);
}
void reallocate_stra(population &pop,population &trial_pop,int stra_idx,int size,double val)
{
int pop_size=pop.size();
int tri_pop_size=trial_pop.size();
if ( pop_size!=tri_pop_size )
{
stringstream ss_err;
ss_err<<"\n"
<<"Error in reallocate_stra:"
<<"pop and trial_pop's size mismatch."
<<"\n";
throw logic_error(ss_err.str());
}
int i;
for ( i=0;i<pop_size;i++ )
{
pop[i].stra[stra_idx].assign(size,val);
trial_pop[i].stra[stra_idx].assign(size,val);
}
}
family_vector niche_ga::select(const population&p,int count)
{
dna_cmeans::u_vector data=fill_data_vector(p);
int cluster_count=static_cast<int>(p.size()*0.10);
dna_cmeans::result r;
if(call_nums>=10){
call_nums=0;
cm.m_center_sequence.erase(cm.m_center_sequence.begin(),cm.m_center_sequence.end());
}
if(cm.m_center_sequence.size()==0){
r=cm.cluster(dna_distance,cluster_count,data.begin(),data.end(),data.size(),1.0);
}
else{
r=cm.cluster(dna_distance,data.begin(),data.end(),data.size());
}
call_nums++;
clusters cls=dna_cmeans::u2clsuters(r.u);
Random rnd;
family_vector result(count);
int c=0; // сколько мы уже отобрали для скрещивания
while(c!=count){
int father_cluster=rnd.uniform(0,cls.size()-1);
if(cls[father_cluster].size()<2)
continue;
int father_index=rnd.uniform(0,cls[father_cluster].size()-1);
int mather_cluster=father_cluster;
int mather_index=rnd.uniform(0,cls[mather_cluster].size()-1);
result[c]=std::make_pair(cls[father_cluster][father_index].first,
cls[mather_cluster][mather_index].first);
c++;
}
return result;
}
std::vector<population::individual_type> best_kill_s_policy::select(population &pop) const
{
pagmo_assert(get_n_individuals(pop) <= pop.size());
// Gets the number of individuals to select
const population::size_type migration_rate = get_n_individuals(pop);
// Create a temporary array of individuals.
std::vector<population::individual_type> result;
// Gets the indexes of the best individuals
std::vector<population::size_type> best_idx = pop.get_best_idx(migration_rate);
// Puts the best individuals in results
for (population::size_type i =0; i< migration_rate; ++i) {
result.push_back(pop.get_individual(best_idx[i]));
}
// Remove them from the original population
// (note: the champion will still carry information on the best guy ...)
for (population::size_type i=0 ; i<migration_rate; ++i) {
pop.reinit(best_idx[i]);
}
return result;
}
// perform (\lambda+\mu) -> (\lambda) selection
void dgea_alg::select(population& pop, population& child_pop)
{
int pop_size=pop.size();
int num_dims=m_ppara->get_dim();
int i;
population pop_merged(pop_size);
vector<vector<double> > prev_x(pop_size);
for ( i=0;i<pop_size;i++ )
{
prev_x[i].resize(num_dims);
pop_merged[i]=pop[i];
// record previous x
prev_x[i]=pop[i].x;
}// for every individual
// push child population at the tail of merged population
copy( child_pop.begin(),child_pop.end(),back_inserter(pop_merged) );
nth_element(pop_merged.begin(), pop_merged.begin()+pop_size, pop_merged.end());
//// heap sorting
//make_heap(pop_merged.begin(),pop_merged.end());
//for ( i=0;i<pop_size;i++ )
//{
// // record delta x for stat purpose
// pop_heap(pop_merged.begin(),pop_merged.end());
// pop_merged.pop_back();
// const individual& new_ind=pop_merged.front();
// pop[i]=new_ind;
// m_alg_stat.delta_x[i]=pop[i].x-prev_x[i];// pop_merged:{parents,offsprings}
//}
for ( i=0;i<pop_size;i++ )
{
pop[i]=pop_merged[i];
// record delta x
m_alg_stat.delta_x[i] = (pop[i].x-prev_x[i]);
}// for every individual
}// end function select
/**
* Updates the constraints scaling vector with the given population.
* @param[in] population pop.
*/
void cstrs_self_adaptive::update_c_scaling(const population &pop)
{
if(*m_original_problem != pop.problem()) {
pagmo_throw(value_error,"The problem linked to the population is not the same as the problem given in argument.");
}
// Let's store some useful variables.
const population::size_type pop_size = pop.size();
// get the constraints dimension
//constraint_vector c(m_original_problem->get_c_dimension(), 0.);
problem::base::c_size_type prob_c_dimension = m_original_problem->get_c_dimension();
problem::base::c_size_type number_of_eq_constraints =
m_original_problem->get_c_dimension() -
m_original_problem->get_ic_dimension();
const std::vector<double> &c_tol = m_original_problem->get_c_tol();
m_c_scaling.resize(m_original_problem->get_c_dimension());
std::fill(m_c_scaling.begin(),m_c_scaling.end(),0.);
// evaluates the scaling factor
for(population::size_type i=0; i<pop_size; i++) {
// updates the current constraint vector
const population::individual_type ¤t_individual = pop.get_individual(i);
const constraint_vector &c = current_individual.cur_c;
// computes scaling with the right definition of the constraints (can be in base problem? currently used
// by con2mo as well)
for(problem::base::c_size_type j=0; j<number_of_eq_constraints; j++) {
m_c_scaling[j] = std::max(m_c_scaling[j], std::max(0., (std::abs(c.at(j)) - c_tol.at(j))) );
}
for(problem::base::c_size_type j=number_of_eq_constraints; j<prob_c_dimension; j++) {
m_c_scaling[j] = std::max(m_c_scaling[j], std::max(0., c.at(j) - c_tol.at(j)) );
}
}
}
bool simpleLSM_job::eval_population_fitness(population &pop)
{
char pwd[FILENAME_MAX];
getcwd(pwd, sizeof(pwd));
log.debug("Current working directory", pwd);
float cputime=0;
float walltime=0;
for (sample &s : pop)
{
prepare_work_dir_for_sample(s);
boost::timer::cpu_timer timer;
eval_sample_fitness(s);
boost::timer::cpu_times elapsed = timer.elapsed();
log.debug(" CPU TIME: ", (elapsed.user + elapsed.system) / 1e9, " seconds", ", WALLCLOCK TIME: ", elapsed.wall / 1e9, " seconds");
cputime+=((elapsed.user + elapsed.system) / 1e9);
walltime+=(elapsed.wall / 1e9);
}
log.debug("total CPU TIME:", cputime, "total WALLCLOCK TIME:", walltime);
log.debug("avergae CPU TIME:", cputime/pop.size(), "avergae WALLCLOCK TIME:", walltime/pop.size());
return true;
}
void snopt::evolve(population &pop) const
{
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const problem::base::size_type D = prob.get_dimension(), prob_i_dimension = prob.get_i_dimension(), prob_c_dimension = prob.get_c_dimension(), prob_f_dimension = prob.get_f_dimension();
const decision_vector &lb = prob.get_lb(), &ub = prob.get_ub();
const population::size_type NP = pop.size();
const problem::base::size_type Dc = D - prob_i_dimension;
const std::vector<double>::size_type D_ineqc = prob.get_ic_dimension();
const std::vector<double>::size_type D_eqc = prob_c_dimension - D_ineqc;
const std::string name = prob.get_name();
//We perform some checks to determine wether the problem/population are suitable for SNOPT
if ( prob_i_dimension != 0 ) {
pagmo_throw(value_error,"No integer part allowed yet....");
}
if ( Dc == 0 ) {
pagmo_throw(value_error,"No continuous part....");
}
if ( prob_f_dimension != 1 ) {
pagmo_throw(value_error,"The problem is not single objective and SNOPT is not suitable to solve it");
}
// Get out if there is nothing to do.
if (NP == 0 || m_major == 0) {
return;
}
// We allocate memory for the decision vector that will be used in the snopt_function_
di_comodo.x.resize(Dc);
di_comodo.c.resize(prob_c_dimension);
di_comodo.f.resize(prob_f_dimension);
// We construct a SnoptProblem_PAGMO passing the pointers to the problem and the allocated
//memory area for the di_comodo vector
snoptProblem_PAGMO SnoptProblem(prob, &di_comodo);
// Allocate and initialize;
integer n = Dc;
// Box-constrained non-linear optimization
integer neF = 1 + prob_c_dimension;
//Memory sizing of A
integer lenA = Dc * (1 + prob_c_dimension); //overestimate
integer *iAfun = new integer[lenA];
integer *jAvar = new integer[lenA];
doublereal *A = new doublereal[lenA];
//Memory sizing of G
int lenG =Dc * (1 + prob_c_dimension); //overestimate
integer *iGfun = new integer[lenG];
integer *jGvar = new integer[lenG];
//Decision vector memory allocation
doublereal *x = new doublereal[n];
doublereal *xlow = new doublereal[n];
doublereal *xupp = new doublereal[n];
doublereal *xmul = new doublereal[n];
integer *xstate = new integer[n];
//Objective function memory allocation
doublereal *F = new doublereal[neF];
doublereal *Flow = new doublereal[neF];
doublereal *Fupp = new doublereal[neF];
doublereal *Fmul = new doublereal[neF];
integer *Fstate = new integer[neF];
integer nxnames = 1;
integer nFnames = 1;
char *xnames = new char[nxnames*8];
char *Fnames = new char[nFnames*8];
integer ObjRow = 0;
doublereal ObjAdd = 0;
// Set the upper and lower bounds. And The initial Guess
int bestidx = pop.get_best_idx();
for (pagmo::problem::base::size_type i = 0; i < Dc; i++) {
xlow[i] = lb[i];
xupp[i] = ub[i];
xstate[i] = 0;
x[i] = pop.get_individual(bestidx).cur_x[i];
}
// Set the bounds for objective, equality and inequality constraints
// 1 - Objective function
Flow[0] = -std::numeric_limits<double>::max();
Fupp[0] = std::numeric_limits<double>::max();
F[0] = pop.get_individual(bestidx).cur_f[0];
// 2 - Equality constraints
for (pagmo::problem::base::size_type i=0; i<D_eqc; ++i) {
Flow[i+1] = 0;
Fupp[i+1] = 0;
}
// 3 - Inequality constraints
for (pagmo::problem::base::size_type i=0; i<D_ineqc; ++i) {
Flow[i+1+D_eqc] = -std::numeric_limits<double>::max();
Fupp[i+1+D_eqc] = 0;
}
// Load the data for SnoptProblem ...
SnoptProblem.setProblemSize( n, neF );
SnoptProblem.setNeG( lenG );
SnoptProblem.setNeA( lenA );
SnoptProblem.setA ( lenA, iAfun, jAvar, A );
SnoptProblem.setG ( lenG, iGfun, jGvar );
SnoptProblem.setObjective ( ObjRow, ObjAdd );
SnoptProblem.setX ( x, xlow, xupp, xmul, xstate );
SnoptProblem.setF ( F, Flow, Fupp, Fmul, Fstate );
SnoptProblem.setXNames ( xnames, nxnames );
SnoptProblem.setFNames ( Fnames, nFnames );
SnoptProblem.setProbName ( name.c_str() ); //This is limited to be 8 characters!!!
SnoptProblem.setUserFun ( snopt_function_ );
//We set some parameters
if (m_screen_output) SnoptProblem.setIntParameter("Summary file",6);
if (m_file_out) SnoptProblem.setPrintFile ( name.c_str() );
SnoptProblem.setIntParameter ( "Derivative option", 0 );
SnoptProblem.setIntParameter ( "Major iterations limit", m_major);
SnoptProblem.setIntParameter ( "Iterations limit",100000);
SnoptProblem.setRealParameter( "Major feasibility tolerance", m_feas);
SnoptProblem.setRealParameter( "Major optimality tolerance", m_opt);
//We set the sparsity structure
int neG;
try
{
std::vector<int> iGfun_vect, jGvar_vect;
prob.set_sparsity(neG,iGfun_vect,jGvar_vect);
for (int i=0; i < neG; i++)
{
iGfun[i] = iGfun_vect[i];
jGvar[i] = jGvar_vect[i];
}
SnoptProblem.setNeG( neG );
SnoptProblem.setNeA( 0 );
SnoptProblem.setG( lenG, iGfun, jGvar );
} //the user did implement the sparsity in the problem
catch (not_implemented_error)
{
SnoptProblem.computeJac();
neG = SnoptProblem.getNeG();
} //the user did not implement the sparsity in the problem
if (m_screen_output)
{
std::cout << "PaGMO 4 SNOPT:" << std::endl << std::endl;
std::cout << "Sparsity pattern set, NeG = " << neG << std::endl;
std::cout << "iGfun: [";
for (int i=0; i<neG-1; ++i) std::cout << iGfun[i] << ",";
std::cout << iGfun[neG-1] << "]" << std::endl;
std::cout << "jGvar: [";
for (int i=0; i<neG-1; ++i) std::cout << jGvar[i] << ",";
std::cout << jGvar[neG-1] << "]" << std::endl;
}
integer Cold = 0;
//HERE WE CALL snoptA routine!!!!!
SnoptProblem.solve( Cold );
//Save the final point making sure it is within the linear bounds
std::copy(x,x+n,di_comodo.x.begin());
decision_vector newx = di_comodo.x;
std::transform(di_comodo.x.begin(), di_comodo.x.end(), pop.get_individual(bestidx).cur_x.begin(), di_comodo.x.begin(),std::minus<double>());
for (integer i=0; i<n; i++)
{
newx[i] = std::min(std::max(lb[i],newx[i]),ub[i]);
}
pop.set_x(bestidx,newx);
pop.set_v(bestidx,di_comodo.x);
//Clean up memory allocated to call the snoptA routine
delete []iAfun;
delete []jAvar;
delete []A;
delete []iGfun;
delete []jGvar;
delete []x;
delete []xlow;
delete []xupp;
delete []xmul;
delete []xstate;
delete []F;
delete []Flow;
delete []Fupp;
delete []Fmul;
delete []Fstate;
delete []xnames;
delete []Fnames;
}
void bee_colony::evolve(population &pop) const
{
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const problem::base::size_type prob_i_dimension = prob.get_i_dimension(), D = prob.get_dimension(), Dc = D - prob_i_dimension, prob_c_dimension = prob.get_c_dimension();
const decision_vector &lb = prob.get_lb(), &ub = prob.get_ub();
const population::size_type NP = (int) pop.size();
//We perform some checks to determine wether the problem/population are suitable for ABC
if ( Dc == 0 ) {
pagmo_throw(value_error,"There is no continuous part in the problem decision vector for ABC to optimise");
}
if ( prob.get_f_dimension() != 1 ) {
pagmo_throw(value_error,"The problem is not single objective and ABC is not suitable to solve it");
}
if ( prob_c_dimension != 0 ) {
pagmo_throw(value_error,"The problem is not box constrained and ABC is not suitable to solve it");
}
if (NP < 2) {
pagmo_throw(value_error,"for ABC at least 2 individuals in the population are needed");
}
// Get out if there is nothing to do.
if (m_iter == 0) {
return;
}
// Some vectors used during evolution are allocated here.
fitness_vector fnew(prob.get_f_dimension());
decision_vector dummy(D,0); //used for initialisation purposes
std::vector<decision_vector > X(NP,dummy); //set of food sources
std::vector<fitness_vector> fit(NP); //food sources fitness
decision_vector temp_solution(D,0);
std::vector<int> trial(NP,0);
std::vector<double> probability(NP);
population::size_type neighbour = 0;
decision_vector::size_type param2change = 0;
std::vector<double> selectionfitness(NP), cumsum(NP), cumsumTemp(NP);
std::vector <population::size_type> selection(NP);
double r = 0;
// Copy the food sources position and their fitness
for ( population::size_type i = 0; i<NP; i++ ) {
X[i] = pop.get_individual(i).cur_x;
fit[i] = pop.get_individual(i).cur_f;
}
// Main ABC loop
for (int j = 0; j < m_iter; ++j) {
//1- Send employed bees
for (population::size_type ii = 0; ii< NP; ++ii) {
//selects a random component (only of the continuous part) of the decision vector
param2change = boost::uniform_int<decision_vector::size_type>(0,Dc-1)(m_urng);
//randomly chose a solution to be used to produce a mutant solution of solution ii
//randomly selected solution must be different from ii
do{
neighbour = boost::uniform_int<population::size_type>(0,NP-1)(m_urng);
}
while(neighbour == ii);
//copy local solution into temp_solution (the whole decision_vector, also the integer part)
for(population::size_type i=0; i<D; ++i) {
temp_solution[i] = X[ii][i];
}
//mutate temp_solution
temp_solution[param2change] = X[ii][param2change] + boost::uniform_real<double>(-1,1)(m_drng) * (X[ii][param2change] - X[neighbour][param2change]);
//if generated parameter value is out of boundaries, it is shifted onto the boundaries*/
if (temp_solution[param2change]<lb[param2change]) {
temp_solution[param2change] = lb[param2change];
}
if (temp_solution[param2change]>ub[param2change]) {
temp_solution[param2change] = ub[param2change];
}
//Calling void prob.objfun(fitness_vector,decision_vector) is more efficient as no memory allocation occur
//A call to fitness_vector prob.objfun(decision_vector) allocates memory for the return value.
prob.objfun(fnew,temp_solution);
//If the new solution is better than the old one replace it with the mutant one and reset its trial counter
if(prob.compare_fitness(fnew, fit[ii])) {
X[ii][param2change] = temp_solution[param2change];
pop.set_x(ii,X[ii]);
prob.objfun(fit[ii], X[ii]); //update the fitness vector
trial[ii] = 0;
}
else {
trial[ii]++; //if the solution can't be improved incrase its trial counter
}
} //End of loop on the population members
//2 - Send onlooker bees
//We scale all fitness values from 0 (worst) to absolute value of the best fitness
fitness_vector worstfit=fit[0];
for (pagmo::population::size_type i = 1; i < NP;i++) {
if (prob.compare_fitness(worstfit,fit[i])) worstfit=fit[i];
}
for (pagmo::population::size_type i = 0; i < NP; i++) {
selectionfitness[i] = fabs(worstfit[0] - fit[i][0]) + 1.;
}
// We build and normalise the cumulative sum
cumsumTemp[0] = selectionfitness[0];
for (pagmo::population::size_type i = 1; i< NP; i++) {
cumsumTemp[i] = cumsumTemp[i - 1] + selectionfitness[i];
}
for (pagmo::population::size_type i = 0; i < NP; i++) {
cumsum[i] = cumsumTemp[i]/cumsumTemp[NP-1];
}
for (pagmo::population::size_type i = 0; i < NP; i++) {
r = m_drng();
for (pagmo::population::size_type j = 0; j < NP; j++) {
if (cumsum[j] > r) {
selection[i]=j;
break;
}
}
}
for(pagmo::population::size_type t = 0; t < NP; ++t) {
r = m_drng();
pagmo::population::size_type ii = selection[t];
//selects a random component (only of the continuous part) of the decision vector
param2change = boost::uniform_int<decision_vector::size_type>(0,Dc-1)(m_urng);
//randomly chose a solution to be used to produce a mutant solution of solution ii
//randomly selected solution must be different from ii
do{
neighbour = boost::uniform_int<population::size_type>(0,NP-1)(m_urng);
}
while(neighbour == ii);
//copy local solution into temp_solution (also integer part)
for(population::size_type i=0; i<D; ++i) {
temp_solution[i] = X[ii][i];
}
//mutate temp_solution
temp_solution[param2change] = X[ii][param2change] + boost::uniform_real<double>(-1,1)(m_drng) * (X[ii][param2change] - X[neighbour][param2change]);
/*if generated parameter value is out of boundaries, it is shifted onto the boundaries*/
if (temp_solution[param2change]<lb[param2change]) {
temp_solution[param2change] = lb[param2change];
}
if (temp_solution[param2change]>ub[param2change]) {
temp_solution[param2change] = ub[param2change];
}
//Calling void prob.objfun(fitness_vector,decision_vector) is more efficient as no memory allocation occur
//A call to fitness_vector prob.objfun(decision_vector) allocates memory for the return value.
prob.objfun(fnew,temp_solution);
//If the new solution is better than the old one replace it with the mutant one and reset its trial counter
if(prob.compare_fitness(fnew, fit[ii])) {
X[ii][param2change] = temp_solution[param2change];
pop.set_x(ii,X[ii]);
prob.objfun(fit[ii], X[ii]); //update the fitness vector
trial[ii] = 0;
}
else {
trial[ii]++; //if the solution can't be improved incrase its trial counter
}
}
//3 - Send scout bees
int maxtrialindex = 0;
for (population::size_type ii=1; ii<NP; ++ii)
{
if (trial[ii] > trial[maxtrialindex]) {
maxtrialindex = ii;
}
}
if(trial[maxtrialindex] >= m_limit)
{
//select a new random solution
for(problem::base::size_type jj = 0; jj < Dc; ++jj) {
X[maxtrialindex][jj] = boost::uniform_real<double>(lb[jj],ub[jj])(m_drng);
}
trial[maxtrialindex] = 0;
pop.set_x(maxtrialindex,X[maxtrialindex]);
}
} // end of main ABC loop
}
/**
* Run the co-evolution algorithm
*
* @param[in,out] pop input/output pagmo::population to be evolved.
*/
void cstrs_co_evolution::evolve(population &pop) const
{
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const population::size_type pop_size = pop.size();
const problem::base::size_type prob_dimension = prob.get_dimension();
// get the constraints dimension
problem::base::c_size_type prob_c_dimension = prob.get_c_dimension();
//We perform some checks to determine wether the problem/population are suitable for co-evolution
if(prob_c_dimension < 1) {
pagmo_throw(value_error,"The problem is not constrained and co-evolution is not suitable to solve it");
}
if(prob.get_f_dimension() != 1) {
pagmo_throw(value_error,"The problem is multiobjective and co-evolution is not suitable to solve it");
}
// Get out if there is nothing to do.
if(pop_size == 0) {
return;
}
//get the dimension of the chromosome of P2
unsigned int pop_2_dim = 0;
switch(m_method)
{
case algorithm::cstrs_co_evolution::SIMPLE:
{
pop_2_dim = 2;
break;
}
case algorithm::cstrs_co_evolution::SPLIT_NEQ_EQ:
{
pop_2_dim = 4;
break;
}
case algorithm::cstrs_co_evolution::SPLIT_CONSTRAINTS:
pop_2_dim = 2*prob.get_c_dimension();
break;
default:
pagmo_throw(value_error,"The constraints co-evolutionary method is not valid.");
break;
}
// split the population into two populations
// the population P1 associated to the modified problem with penalized fitness
// and population P2 encoding the penalty weights
// Populations size
population::size_type pop_1_size = pop_size;
population::size_type pop_2_size = m_pop_penalties_size;
//Creates problem associated to P2
problem::cstrs_co_evolution_penalty prob_2(prob,pop_2_dim,pop_2_size);
prob_2.set_bounds(m_pen_lower_bound,m_pen_upper_bound);
//random initialization of the P2 chromosome (needed for the fist generation)
std::vector<decision_vector> pop_2_x(pop_2_size);
std::vector<fitness_vector> pop_2_f(pop_2_size);
for(population::size_type j=0; j<pop_2_size; j++) {
pop_2_x[j] = decision_vector(pop_2_dim,0.);
// choose random coefficients between lower bound and upper bound
for(population::size_type i=0; i<pop_2_dim;i++) {
pop_2_x[j][i] = boost::uniform_real<double>(m_pen_lower_bound,m_pen_upper_bound)(m_drng);
}
}
//vector of the population P1. Initialized with clones of the original population
std::vector<population> pop_1_vector;
for(population::size_type i=0; i<pop_2_size; i++){
pop_1_vector.push_back(population(pop));
}
// Main Co-Evolution loop
for(int k=0; k<m_gen; k++) {
// for each individuals of pop 2, evolve the current population,
// and store the position of the feasible idx
for(population::size_type j=0; j<pop_2_size; j++) {
problem::cstrs_co_evolution prob_1(prob, pop_1_vector.at(j), m_method);
// modify the problem by setting decision vector encoding penalty
// coefficients w1 and w2 in prob 1
prob_1.set_penalty_coeff(pop_2_x.at(j));
// creating the POPULATION 1 instance based on the
// updated prob 1
// prob_1 is a BASE_META???? THE CLONE OF prob_1 IN POP_1 IS AT THE LEVEL OF
// THE BASE CLASS AND NOT AT THE LEVEL OF THE BASE_META, NO?!?
population pop_1(prob_1,0);
// initialize P1 chromosomes. The fitnesses related to problem 1 are computed
for(population::size_type i=0; i<pop_1_size; i++) {
pop_1.push_back(pop_1_vector.at(j).get_individual(i).cur_x);
}
// evolve the P1 instance
m_original_algo->evolve(pop_1);
//updating the original problem population (computation of fitness and constraints)
pop_1_vector.at(j).clear();
for(population::size_type i=0; i<pop_1_size; i++){
pop_1_vector.at(j).push_back(pop_1.get_individual(i).cur_x);
}
// set up penalization variables needs for the population 2
// the constraints has not been evaluated yet.
prob_2.update_penalty_coeff(j,pop_2_x.at(j),pop_1_vector.at(j));
}
// creating the POPULATION 2 instance based on the
// updated prob 2
population pop_2(prob_2,0);
// compute the fitness values of the second population
for(population::size_type i=0; i<pop_2_size; i++) {
pop_2.push_back(pop_2_x[i]);
}
m_original_algo_penalties->evolve(pop_2);
// store the new chromosomes
for(population::size_type i=0; i<pop_2_size; i++) {
pop_2_x[i] = pop_2.get_individual(i).cur_x;
pop_2_f[i] = pop_2.get_individual(i).cur_f;
}
// Check the exit conditions (every 40 generations, just as DE)
if(k % 40 == 0) {
// finds the best population
population::size_type best_idx = 0;
for(population::size_type j=1; j<pop_2_size; j++) {
if(pop_2_f[j][0] < pop_2_f[best_idx][0]) {
best_idx = j;
}
}
const population ¤t_population = pop_1_vector.at(best_idx);
decision_vector tmp(prob_dimension);
double dx = 0;
for(decision_vector::size_type i=0; i<prob_dimension; i++) {
tmp[i] = current_population.get_individual(current_population.get_worst_idx()).best_x[i] -
current_population.get_individual(current_population.get_best_idx()).best_x[i];
dx += std::fabs(tmp[i]);
}
if(dx < m_xtol ) {
if (m_screen_output) {
std::cout << "Exit condition -- xtol < " << m_xtol << std::endl;
}
break;
}
double mah = std::fabs(current_population.get_individual(current_population.get_worst_idx()).best_f[0] -
current_population.get_individual(current_population.get_best_idx()).best_f[0]);
if(mah < m_ftol) {
if(m_screen_output) {
std::cout << "Exit condition -- ftol < " << m_ftol << std::endl;
}
break;
}
// outputs current values
if(m_screen_output) {
std::cout << "Generation " << k << " ***" << std::endl;
std::cout << " Best global fitness: " << current_population.champion().f << std::endl;
std::cout << " xtol: " << dx << ", ftol: " << mah << std::endl;
std::cout << " xtol: " << dx << ", ftol: " << mah << std::endl;
}
}
}
// find the best fitness population in the final pop
population::size_type best_idx = 0;
for(population::size_type j=1; j<pop_2_size; j++) {
if(pop_2_f[j][0] < pop_2_f[best_idx][0]) {
best_idx = j;
}
}
// store the final population in the main population
// can't avoid to recompute the vectors here, otherwise we
// clone the problem stored in the population with the
// pop = operator!
pop.clear();
for(population::size_type i=0; i<pop_1_size; i++) {
pop.push_back(pop_1_vector.at(best_idx).get_individual(i).cur_x);
}
}
population choosiness_breeder::breed(const population &generation, const std::vector<double> &fitnesses) const
{
using namespace std;
population ret;
tree_crossover_reproducer repr;
node_crossover_reproducer mrepr;
builder random_builder;
deque<const agent *> mut_gen;
for(const auto &a : generation) mut_gen.push_back(&a);
experimental_parameters e = exp_params();
const uint32_t growth_tree_min = e["growth_tree_min"];
const uint32_t growth_tree_max = e["growth_tree_max"];
const double max_dist = sqrt(_maze.rows() * _maze.rows() + _maze.columns() * _maze.columns());
_phenotypes.clear();
for(const auto &a : mut_gen) _phenotypes.at(a->final_state().row, a->final_state().col).push_back(a);
map<const agent *, vector<const agent *> > matches;
for(auto ait = mut_gen.begin(); ait != mut_gen.end();)
{
const agent *const a = *ait;
const vector<vector<const agent *> > plausible_mates =
_phenotypes.find_in_range(a->final_state().row, a->final_state().col,
a->chromosomes()[0]);
vector<const agent *> flattened_mates;
for(const auto &v : plausible_mates) flattened_mates.insert(flattened_mates.end(), v.begin(), v.end());
for(auto it = flattened_mates.begin(); it != flattened_mates.end();)
{
if(distance((*it)->final_state().col, (*it)->final_state().row,
a->final_state().col, a->final_state().row) > (*it)->chromosomes()[0])
{
it = flattened_mates.erase(it);
continue;
}
++it;
}
if(flattened_mates.empty())
{
ait = mut_gen.erase(ait);
continue;
}
matches[a] = flattened_mates;
++ait;
}
const population::size_type new_size = generation.size();
random_shuffle(mut_gen.begin(), mut_gen.end());
while(ret.size() < new_size)
{
for(const auto &mother : mut_gen)
{
if(ret.size() >= new_size) break;
const vector<const agent *> &fathers = matches[mother];
if(fathers.empty()) continue;
const agent &father = *fathers[rand() % fathers.size()];
agent mutant = random_builder.grow(program_types_set, growth_tree_min, growth_tree_max);
mutant.set_chromosomes(vector<double> { rand_normal() * max_dist });
vector<agent> children = repr.reproduce(vector<agent> { *mother, father });
if(children.empty()) continue;
agent &child = children[0];
child.set_chromosomes(vector<double> { (mother->chromosomes()[0] + father.chromosomes()[0]) / 2.0 });
agent res = mrepr.reproduce(vector<agent> { child, mutant })[0];
res.set_chromosomes(vector<double> { (child.chromosomes()[0] + mutant.chromosomes()[0]) / 2.0 });
ret.push_back(res);
}
}
return ret;
}
/**
* Updates the penalty coefficients with the given population.
* @param[in] population pop.
*/
void cstrs_self_adaptive::update_penalty_coeff(const population &pop)
{
if(*m_original_problem != pop.problem()) {
pagmo_throw(value_error,"The problem linked to the population is not the same as the problem given in argument.");
}
// Let's store some useful variables.
const population::size_type pop_size = pop.size();
// Get out if there is nothing to do.
if (pop_size == 0) {
return;
}
m_map_fitness.clear();
m_map_constraint.clear();
// store f and c in maps depending on the the x hash
for(population::size_type i=0; i<pop_size; i++) {
const population::individual_type ¤t_individual = pop.get_individual(i);
// m_map_fitness.insert(std::pair<std::size_t, fitness_vector>(m_decision_vector_hash(current_individual.cur_x),current_individual.cur_f));
m_map_fitness[m_decision_vector_hash(current_individual.cur_x)]=current_individual.cur_f;
m_map_constraint[m_decision_vector_hash(current_individual.cur_x)]=current_individual.cur_c;
}
std::vector<population::size_type> feasible_idx(0);
std::vector<population::size_type> infeasible_idx(0);
// store indexes of feasible and non feasible individuals
for(population::size_type i=0; i<pop_size; i++) {
const population::individual_type ¤t_individual = pop.get_individual(i);
if(m_original_problem->feasibility_c(current_individual.cur_c)) {
feasible_idx.push_back(i);
} else {
infeasible_idx.push_back(i);
}
}
// if the population is only feasible, then nothing is done
if(infeasible_idx.size() == 0) {
update_c_scaling(pop);
m_apply_penalty_1 = false;
m_scaling_factor = 0.;
return;
}
m_apply_penalty_1 = false;
m_scaling_factor = 0.;
// updates the c_scaling, needed for solution infeasibility computation
update_c_scaling(pop);
// evaluate solutions infeasibility
//compute_pop_solution_infeasibility(solution_infeasibility, pop);
std::vector<double> solution_infeasibility(pop_size);
std::fill(solution_infeasibility.begin(),solution_infeasibility.end(),0.);
// evaluate solutions infeasibility
solution_infeasibility.resize(pop_size);
std::fill(solution_infeasibility.begin(),solution_infeasibility.end(),0.);
for(population::size_type i=0; i<pop_size; i++) {
const population::individual_type ¤t_individual = pop.get_individual(i);
// compute the infeasibility of the constraint
solution_infeasibility[i] = compute_solution_infeasibility(current_individual.cur_c);
}
// search position of x_hat_down, x_hat_up and x_hat_round
population::size_type hat_down_idx = -1;
population::size_type hat_up_idx = -1;
population::size_type hat_round_idx = -1;
// first case, the population contains at least one feasible solution
if(feasible_idx.size() > 0) {
// initialize hat_down_idx
hat_down_idx = feasible_idx.at(0);
// x_hat_down = feasible individual with lowest objective value in p
for(population::size_type i=0; i<feasible_idx.size(); i++) {
const population::size_type current_idx = feasible_idx.at(i);
const population::individual_type ¤t_individual = pop.get_individual(current_idx);
if(m_original_problem->compare_fitness(current_individual.cur_f, pop.get_individual(hat_down_idx).cur_f)) {
hat_down_idx = current_idx;
}
}
// hat down is now available
fitness_vector f_hat_down = pop.get_individual(hat_down_idx).cur_f;
// x_hat_up value depends if the population contains infeasible individual with objective
// function better than f_hat_down
bool pop_contains_infeasible_f_better_x_hat_down = false;
for(population::size_type i=0; i<infeasible_idx.size(); i++) {
const population::size_type current_idx = infeasible_idx.at(i);
const population::individual_type ¤t_individual = pop.get_individual(current_idx);
if(m_original_problem->compare_fitness(current_individual.cur_f, f_hat_down)) {
pop_contains_infeasible_f_better_x_hat_down = true;
// initialize hat_up_idx
hat_up_idx = current_idx;
break;
}
}
if(pop_contains_infeasible_f_better_x_hat_down) {
// hat_up_idx is already initizalized
// gets the individual with maximum infeasibility and objfun lower than f_hat_down
for(population::size_type i=0; i<infeasible_idx.size(); i++) {
const population::size_type current_idx = infeasible_idx.at(i);
const population::individual_type ¤t_individual = pop.get_individual(current_idx);
if(m_original_problem->compare_fitness(current_individual.cur_f, f_hat_down) &&
(solution_infeasibility.at(current_idx) >= solution_infeasibility.at(hat_up_idx)) ) {
if(solution_infeasibility.at(current_idx) == solution_infeasibility.at(hat_up_idx)) {
if(m_original_problem->compare_fitness(current_individual.cur_f, pop.get_individual(hat_up_idx).cur_f)) {
hat_up_idx = current_idx;
}
} else {
hat_up_idx = current_idx;
}
}
}
// apply penalty 1
m_apply_penalty_1 = true;
} else {
// all the infeasible soutions have an objective function value greater than f_hat_down
// the worst is the one that has the maximum infeasibility
// initialize hat_up_idx
hat_up_idx = infeasible_idx.at(0);
for(population::size_type i=0; i<infeasible_idx.size(); i++) {
const population::size_type current_idx = infeasible_idx.at(i);
const population::individual_type ¤t_individual = pop.get_individual(current_idx);
if(solution_infeasibility.at(current_idx) >= solution_infeasibility.at(hat_up_idx)) {
if(solution_infeasibility.at(current_idx) == solution_infeasibility.at(hat_up_idx)) {
if(m_original_problem->compare_fitness(pop.get_individual(hat_up_idx).cur_f, current_individual.cur_f)) {
hat_up_idx = current_idx;
}
} else {
hat_up_idx = current_idx;
}
}
}
// do not apply penalty 1
m_apply_penalty_1 = false;
}
} else { // case where there is no feasible solution in the population
// best is the individual with the lowest infeasibility
hat_down_idx = 0;
hat_up_idx = 0;
for(population::size_type i=0; i<pop_size; i++) {
const population::individual_type ¤t_individual = pop.get_individual(i);
if(solution_infeasibility.at(i) <= solution_infeasibility.at(hat_down_idx)) {
if(solution_infeasibility.at(i) == solution_infeasibility.at(hat_down_idx)) {
if(m_original_problem->compare_fitness(current_individual.cur_f, pop.get_individual(hat_down_idx).cur_f)) {
hat_down_idx = i;
}
} else {
hat_down_idx = i;
}
}
}
// worst individual
for(population::size_type i=0; i<pop_size; i++) {
const population::individual_type ¤t_individual = pop.get_individual(i);
if(solution_infeasibility.at(i) >= solution_infeasibility.at(hat_up_idx)) {
if(solution_infeasibility.at(i) == solution_infeasibility.at(hat_up_idx)) {
if(m_original_problem->compare_fitness(pop.get_individual(hat_up_idx).cur_f, current_individual.cur_f)) {
hat_up_idx = i;
}
} else {
hat_up_idx = i;
}
}
}
// apply penalty 1 to the population
m_apply_penalty_1 = true;
}
// stores the hat round idx, i.e. the solution with highest objective
// function value in the population
hat_round_idx = 0;
for(population::size_type i=0; i<pop_size; i++) {
const population::individual_type ¤t_individual = pop.get_individual(i);
if(m_original_problem->compare_fitness(pop.get_individual(hat_round_idx).cur_f, current_individual.cur_f)) {
hat_round_idx = i;
}
}
// get the objective function values of the three individuals
m_f_hat_round = pop.get_individual(hat_round_idx).cur_f;
m_f_hat_down = pop.get_individual(hat_down_idx).cur_f;
m_f_hat_up = pop.get_individual(hat_up_idx).cur_f;
// get the solution infeasibility values of the three individuals
m_i_hat_round = solution_infeasibility.at(hat_round_idx);
m_i_hat_down = solution_infeasibility.at(hat_down_idx);
m_i_hat_up = solution_infeasibility.at(hat_up_idx);
// computes the scaling factor
m_scaling_factor = 0.;
// evaluates scaling factor
if(m_original_problem->compare_fitness(m_f_hat_down, m_f_hat_up)) {
m_scaling_factor = (m_f_hat_round[0] - m_f_hat_up[0]) / m_f_hat_up[0];
} else {
m_scaling_factor = (m_f_hat_round[0] - m_f_hat_down[0]) / m_f_hat_down[0];
}
if(m_f_hat_up[0] == m_f_hat_round[0]) {
m_scaling_factor = 0.;
}
}
/**
* The best member of the population will be used as starting point for the minimisation process. The algorithm will stop
* if the gradient falls below the grad_tol parameter, if the maximum number of iterations max_iter is exceeded or if
* the inner GSL routine call reports an error (which will be logged on std::cout). After the end of the minimisation process,
* the minimised decision vector will replace the best individual in the population, after being modified to fall within
* the problem bounds if necessary.
*
* @param[in,out] pop population to evolve.
*/
void gsl_gradient::evolve(population &pop) const
{
// Do nothing if the population is empty.
if (!pop.size()) {
return;
}
// Useful variables.
const problem::base &problem = pop.problem();
if (problem.get_f_dimension() != 1) {
pagmo_throw(value_error,"this algorithm does not support multi-objective optimisation");
}
if (problem.get_c_dimension()) {
pagmo_throw(value_error,"this algorithm does not support constrained optimisation");
}
const problem::base::size_type cont_size = problem.get_dimension() - problem.get_i_dimension();
if (!cont_size) {
pagmo_throw(value_error,"the problem has no continuous part");
}
// Extract the best individual.
const population::size_type best_ind_idx = pop.get_best_idx();
const population::individual_type &best_ind = pop.get_individual(best_ind_idx);
// GSL wrapper parameters structure.
objfun_wrapper_params params;
params.p = &problem;
// Integer part of the temporay decision vector must be filled with the integer part of the best individual,
// which will not be optimised.
params.x.resize(problem.get_dimension());
std::copy(best_ind.cur_x.begin() + cont_size, best_ind.cur_x.end(), params.x.begin() + cont_size);
params.f.resize(1);
params.step_size = m_numdiff_step_size;
// GSL function structure.
gsl_multimin_function_fdf gsl_func;
gsl_func.n = boost::numeric_cast<std::size_t>(cont_size);
gsl_func.f = &objfun_wrapper;
gsl_func.df = &d_objfun_wrapper;
gsl_func.fdf = &fd_objfun_wrapper;
gsl_func.params = (void *)¶ms;
// Minimiser.
gsl_multimin_fdfminimizer *s = 0;
// This will be the starting point.
gsl_vector *x = 0;
// Here we start the allocations.
// Recast as size_t here, in order to avoid potential overflows later.
const std::size_t s_cont_size = boost::numeric_cast<std::size_t>(cont_size);
// Allocate and check the allocation results.
x = gsl_vector_alloc(s_cont_size);
const gsl_multimin_fdfminimizer_type *minimiser = get_gsl_minimiser_ptr();
pagmo_assert(minimiser);
s = gsl_multimin_fdfminimizer_alloc(minimiser,s_cont_size);
// Check the allocations.
check_allocs(x,s);
// Fill in the starting point (from the best individual).
for (std::size_t i = 0; i < s_cont_size; ++i) {
gsl_vector_set(x,i,best_ind.cur_x[i]);
}
// Init the solver.
gsl_multimin_fdfminimizer_set(s,&gsl_func,x,m_step_size,m_tol);
// Iterate.
std::size_t iter = 0;
int status;
try {
do
{
++iter;
status = gsl_multimin_fdfminimizer_iterate(s);
if (status) {
break;
}
status = gsl_multimin_test_gradient(s->gradient,m_grad_tol);
} while (status == GSL_CONTINUE && iter < m_max_iter);
} catch (const std::exception &e) {
// Cleanup and re-throw.
cleanup(x,s);
throw e;
} catch (...) {
// Cleanup and throw.
cleanup(x,s);
pagmo_throw(std::runtime_error,"unknown exception caught in gsl_gradient::evolve");
}
// Free up resources.
cleanup(x,s);
// Check the generated individual and change it to respect the bounds as necessary.
for (problem::base::size_type i = 0; i < cont_size; ++i) {
if (params.x[i] < problem.get_lb()[i]) {
params.x[i] = problem.get_lb()[i];
}
if (params.x[i] > problem.get_ub()[i]) {
params.x[i] = problem.get_ub()[i];
}
}
// Replace the best individual.
pop.set_x(best_ind_idx,params.x);
}
void sa_corana::evolve(population &pop) const {
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const problem::base::size_type D = prob.get_dimension(), prob_i_dimension = prob.get_i_dimension(), prob_c_dimension = prob.get_c_dimension(), prob_f_dimension = prob.get_f_dimension();
const decision_vector &lb = prob.get_lb(), &ub = prob.get_ub();
const population::size_type NP = pop.size();
const problem::base::size_type Dc = D - prob_i_dimension;
//We perform some checks to determine wether the problem/population are suitable for sa_corana
if ( Dc == 0 ) {
pagmo_throw(value_error,"There is no continuous part in the problem decision vector for sa_corana to optimise");
}
if ( prob_c_dimension != 0 ) {
pagmo_throw(value_error,"The problem is not box constrained and sa_corana is not suitable to solve it");
}
if ( prob_f_dimension != 1 ) {
pagmo_throw(value_error,"The problem is not single objective and sa_corana is not suitable to solve it");
}
//Determines the number of temperature adjustment for the annealing procedure
const size_t n_T = m_niter / (m_step_adj * m_bin_size * Dc);
// Get out if there is nothing to do.
if (NP == 0 || m_niter == 0) {
return;
}
if (n_T == 0) {
pagmo_throw(value_error,"n_T is zero, increase niter");
}
//Starting point is the best individual
const int bestidx = pop.get_best_idx();
const decision_vector &x0 = pop.get_individual(bestidx).cur_x;
const fitness_vector &fit0 = pop.get_individual(bestidx).cur_f;
//Determines the coefficient to dcrease the temperature
const double Tcoeff = std::pow(m_Tf/m_Ts,1.0/(double)(n_T));
//Stores the current and new points
decision_vector xNEW = x0, xOLD = xNEW;
fitness_vector fNEW = fit0, fOLD = fNEW;
//Stores the adaptive steps of each component (integer part included but not used)
decision_vector step(D,m_range);
//Stores the number of accepted points per component (integer part included but not used)
std::vector<int> acp(D,0) ;
double ratio = 0, currentT = m_Ts, probab = 0;
//Main SA loops
for (size_t jter = 0; jter < n_T; ++jter) {
for (int mter = 0; mter < m_step_adj; ++mter) {
for (int kter = 0; kter < m_bin_size; ++kter) {
size_t nter = boost::uniform_int<int>(0,Dc-1)(m_urng);
for (size_t numb = 0; numb < Dc ; ++numb) {
nter = (nter + 1) % Dc;
//We modify the current point actsol by mutating its nter component within
//a step that we will later adapt
xNEW[nter] = xOLD[nter] + boost::uniform_real<double>(-1,1)(m_drng) * step[nter] * (ub[nter]-lb[nter]);
// If new solution produced is infeasible ignore it
if ((xNEW[nter] > ub[nter]) || (xNEW[nter] < lb[nter])) {
xNEW[nter]=xOLD[nter];
continue;
}
//And we valuate the objective function for the new point
prob.objfun(fNEW,xNEW);
// We decide wether to accept or discard the point
if (prob.compare_fitness(fNEW,fOLD) ) {
//accept
xOLD[nter] = xNEW[nter];
fOLD = fNEW;
acp[nter]++; //Increase the number of accepted values
} else {
//test it with Boltzmann to decide the acceptance
probab = exp ( - fabs(fOLD[0] - fNEW[0] ) / currentT );
// we compare prob with a random probability.
if (probab > m_drng()) {
xOLD[nter] = xNEW[nter];
fOLD = fNEW;
acp[nter]++; //Increase the number of accepted values
} else {
xNEW[nter] = xOLD[nter];
}
} // end if
} // end for(nter = 0; ...
} // end for(kter = 0; ...
// adjust the step (adaptively)
for (size_t iter = 0; iter < Dc; ++iter) {
ratio = (double)acp[iter]/(double)m_bin_size;
acp[iter] = 0; //reset the counter
if (ratio > .6) {
//too many acceptances, increase the step by a factor 3 maximum
step[iter] = step [iter] * (1 + 2 *(ratio - .6)/.4);
} else {
if (ratio < .4) {
//too few acceptance, decrease the step by a factor 3 maximum
step [iter]= step [iter] / (1 + 2 * ((.4 - ratio)/.4));
};
};
//And if it becomes too large, reset it to its initial value
if ( step[iter] > m_range ) {
step [iter] = m_range;
};
}
}
// Cooling schedule
currentT *= Tcoeff;
}
if ( prob.compare_fitness(fOLD,fit0) ){
pop.set_x(bestidx,xOLD); //new evaluation is possible here......
std::transform(xOLD.begin(), xOLD.end(), pop.get_individual(bestidx).cur_x.begin(), xOLD.begin(),std::minus<double>());
pop.set_v(bestidx,xOLD);
}
}
void ihs::evolve(population &pop) const
{
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const problem::base::size_type prob_dimension = prob.get_dimension(), prob_i_dimension = prob.get_i_dimension();
const decision_vector &lb = prob.get_lb(), &ub = prob.get_ub();
const population::size_type pop_size = pop.size();
// Get out if there is nothing to do.
if (pop_size == 0 || m_gen == 0) {
return;
}
decision_vector lu_diff(prob_dimension);
for (problem::base::size_type i = 0; i < prob_dimension; ++i) {
lu_diff[i] = ub[i] - lb[i];
}
// Int distribution to be used when picking random individuals.
boost::uniform_int<population::size_type> uni_int(0,pop_size - 1);
const double c = std::log(m_bw_min/m_bw_max) / m_gen;
// Temporary individual used during evolution.
population::individual_type tmp;
tmp.cur_x.resize(prob_dimension);
tmp.cur_f.resize(prob.get_f_dimension());
tmp.cur_c.resize(prob.get_c_dimension());
for (std::size_t g = 0; g < m_gen; ++g) {
const double ppar_cur = m_ppar_min + ((m_ppar_max - m_ppar_min) * g) / m_gen, bw_cur = m_bw_max * std::exp(c * g);
// Continuous part.
for (problem::base::size_type i = 0; i < prob_dimension - prob_i_dimension; ++i) {
if (m_drng() < m_phmcr) {
// tmp's i-th chromosome element is the one from a randomly chosen individual.
tmp.cur_x[i] = pop.get_individual(uni_int(m_urng)).cur_x[i];
// Do pitch adjustment with ppar_cur probability.
if (m_drng() < ppar_cur) {
// Randomly, add or subtract pitch from the current chromosome element.
if (m_drng() > .5) {
tmp.cur_x[i] += m_drng() * bw_cur * lu_diff[i];
} else {
tmp.cur_x[i] -= m_drng() * bw_cur * lu_diff[i];
}
// Handle the case in which we added or subtracted too much and ended up out
// of boundaries.
if (tmp.cur_x[i] > ub[i]) {
tmp.cur_x[i] = boost::uniform_real<double>(lb[i],ub[i])(m_drng);
} else if (tmp.cur_x[i] < lb[i]) {
tmp.cur_x[i] = boost::uniform_real<double>(lb[i],ub[i])(m_drng);
}
}
} else {
// Pick randomly within the bounds.
tmp.cur_x[i] = boost::uniform_real<double>(lb[i],ub[i])(m_drng);
}
}
//Integer Part
for (problem::base::size_type i = prob_dimension - prob_i_dimension; i < prob_dimension; ++i) {
if (m_drng() < m_phmcr) {
tmp.cur_x[i] = pop.get_individual(uni_int(m_urng)).cur_x[i];
if (m_drng() < ppar_cur) {
if (m_drng() > .5) {
tmp.cur_x[i] += double_to_int::convert(m_drng() * bw_cur * lu_diff[i]);
} else {
tmp.cur_x[i] -= double_to_int::convert(m_drng() * bw_cur * lu_diff[i]);
}
// Wrap over in case we went past the bounds.
if (tmp.cur_x[i] > ub[i]) {
tmp.cur_x[i] = lb[i] + double_to_int::convert(tmp.cur_x[i] - ub[i]) % static_cast<int>(lu_diff[i]);
} else if (tmp.cur_x[i] < lb[i]) {
tmp.cur_x[i] = ub[i] - double_to_int::convert(lb[i] - tmp.cur_x[i]) % static_cast<int>(lu_diff[i]);
}
}
} else {
// Pick randomly within the bounds.
tmp.cur_x[i] = boost::uniform_int<int>(lb[i],ub[i])(m_urng);
}
}
// And we push him back
pop.push_back(tmp.cur_x);
// We locate the worst individual.
const population::size_type worst_idx = pop.get_worst_idx();
// And we get rid of him :)
pop.erase(worst_idx);
}
}
void ipopt::evolve(population &pop) const
{
// Let's store some useful variables.
const problem::base &prob = pop.problem();
const problem::base::size_type D = prob.get_dimension(), prob_i_dimension = prob.get_i_dimension(), prob_f_dimension = prob.get_f_dimension();
const population::size_type NP = pop.size();
const problem::base::size_type Dc = D - prob_i_dimension;
const std::string name = prob.get_name();
//We perform some checks to determine wether the problem/population are suitable for IPOPT
if ( prob_i_dimension != 0 ) {
pagmo_throw(value_error,"No integer part allowed yet....");
}
if ( Dc == 0 ) {
pagmo_throw(value_error,"No continuous part....");
}
if ( prob_f_dimension != 1 ) {
pagmo_throw(value_error,"The problem is not single objective and IPOPT is not suitable to solve it");
}
// Get out if there is nothing to do.
if (NP == 0 || m_max_iter == 0) {
return;
}
//create an instance of the ipopt_problem
::Ipopt::SmartPtr< ::Ipopt::TNLP> pagmo_nlp = new ipopt_problem(&pop);
//create an instance of the IpoptApplication
::Ipopt::SmartPtr< ::Ipopt::IpoptApplication> m_app = new ::Ipopt::IpoptApplication(m_screen_output,false);
if (!m_nlp_scaling_method) {
m_app->Options()->SetStringValue("nlp_scaling_method", "none");
}
m_app->Options()->SetStringValue("hessian_approximation", "limited-memory");
m_app->Options()->SetIntegerValue("print_level", 5);
// Termination Criteria 1: iterations
m_app->Options()->SetIntegerValue("max_iter", m_max_iter);
// Termination Criteria 2: tolerance
m_app->Options()->SetNumericValue("tol", 1.);
m_app->Options()->SetNumericValue("dual_inf_tol", m_dual_inf_tol);
m_app->Options()->SetNumericValue("constr_viol_tol", m_constr_viol_tol);
m_app->Options()->SetNumericValue("compl_inf_tol", m_compl_inf_tol);
m_app->Options()->SetNumericValue("obj_scaling_factor", m_obj_scaling_factor);
m_app->Options()->SetNumericValue("mu_init", m_mu_init);
// Intialize the IpoptApplication and process the options
Ipopt::ApplicationReturnStatus status;
status = m_app->Initialize();
if (status != Ipopt::Solve_Succeeded) {
pagmo_throw(value_error, "Error during IPOPT initialization!");
}
// Ask Ipopt to solve the problem
status = m_app->OptimizeTNLP(pagmo_nlp);
}
/**
* Runs the Inverover algorithm for the number of generations specified in the constructor.
*
* @param[in,out] pop input/output pagmo::population to be evolved.
*/
void inverover::evolve(population &pop) const
{
const problem::base_tsp* prob;
//check if problem is of type pagmo::problem::base_tsp
try {
const problem::base_tsp& tsp_prob = dynamic_cast<const problem::base_tsp &>(pop.problem());
prob = &tsp_prob;
}
catch (const std::bad_cast& e) {
pagmo_throw(value_error,"Problem not of type pagmo::problem::base_tsp");
}
// Let's store some useful variables.
const population::size_type NP = pop.size();
const problem::base::size_type Nv = prob->get_n_cities();
// Initializing the random number generators
boost::uniform_real<double> uniform(0.0, 1.0);
boost::variate_generator<boost::lagged_fibonacci607 &, boost::uniform_real<double> > unif_01(m_drng, uniform);
boost::uniform_int<int> NPless1(0, NP - 2);
boost::variate_generator<boost::mt19937 &, boost::uniform_int<int> > unif_NPless1(m_urng, NPless1);
boost::uniform_int<int> Nv_(0, Nv - 1);
boost::variate_generator<boost::mt19937 &, boost::uniform_int<int> > unif_Nv(m_urng, Nv_);
boost::uniform_int<int> Nvless1(0, Nv - 2);
boost::variate_generator<boost::mt19937 &, boost::uniform_int<int> > unif_Nvless1(m_urng, Nvless1);
//create own local population
std::vector<decision_vector> my_pop(NP, decision_vector(Nv));
//check if some individuals in the population that is passed as a function input are feasible.
bool feasible;
std::vector<int> not_feasible;
for (size_t i = 0; i < NP; i++) {
feasible = prob->feasibility_x(pop.get_individual(i).cur_x);
if(feasible) { //if feasible store it in my_pop
switch(prob->get_encoding()) {
case problem::base_tsp::FULL:
my_pop[i] = prob->full2cities(pop.get_individual(i).cur_x);
break;
case problem::base_tsp::RANDOMKEYS:
my_pop[i] = prob->randomkeys2cities(pop.get_individual(i).cur_x);
break;
case problem::base_tsp::CITIES:
my_pop[i] = pop.get_individual(i).cur_x;
break;
}
} else {
not_feasible.push_back(i);
}
}
//replace the not feasible individuals by feasible ones
int i;
switch (m_ini_type) {
case 0:
{
//random initialization (produces feasible individuals)
for (size_t ii = 0; ii < not_feasible.size(); ii++) {
i = not_feasible[ii];
for (size_t j = 0; j < Nv; j++) {
my_pop[i][j] = j;
}
}
int tmp;
size_t rnd_idx;
for (size_t j = 1; j < Nv-1; j++) {
boost::uniform_int<int> dist_(j, Nv - 1);
boost::variate_generator<boost::mt19937 &, boost::uniform_int<int> > dist(m_urng,dist_);
for (size_t ii = 0; ii < not_feasible.size(); ii++) {
i = not_feasible[ii];
rnd_idx = dist();
tmp = my_pop[i][j];
my_pop[i][j] = my_pop[i][rnd_idx];
my_pop[i][rnd_idx] = tmp;
}
}
break;
}
case 1:
{
//initialize with nearest neighbor algorithm
std::vector<int> starting_notes(std::max(Nv,not_feasible.size()));
for (size_t j = 0; j < starting_notes.size(); j++) {
starting_notes[j] = j;
}
//std::shuffle(starting_notes.begin(), starting_notes.end(), m_urng);
for (size_t ii = 0; ii < not_feasible.size(); ii++) {
i = not_feasible[ii];
pagmo::population one_ind_pop(pop.problem(), 1);
std::cout << starting_notes[i] << ' ';
pagmo::algorithm::nn_tsp algo(starting_notes[i] % Nv);
algo.evolve(one_ind_pop);
switch( prob->get_encoding() ) {
case problem::base_tsp::FULL:
my_pop[i] = prob->full2cities(one_ind_pop.get_individual(0).cur_x);
break;
case problem::base_tsp::RANDOMKEYS:
my_pop[i] = prob->randomkeys2cities(one_ind_pop.get_individual(0).cur_x);
break;
case problem::base_tsp::CITIES:
my_pop[i] = one_ind_pop.get_individual(0).cur_x;
break;
}
std::cout << i << ' ' << one_ind_pop.get_individual(0).cur_f << std::endl;
}
break;
}
default:
pagmo_throw(value_error,"Invalid initialization type");
}
std::vector<fitness_vector> fitness(NP, fitness_vector(1));
for(size_t i=0; i < NP; i++){
switch( prob->get_encoding() ) {
case problem::base_tsp::FULL:
fitness[i] = prob->objfun(prob->full2cities(my_pop[i]));
break;
case problem::base_tsp::RANDOMKEYS:
fitness[i] = prob->objfun(prob->cities2randomkeys(my_pop[i], pop.get_individual(i).cur_x));
break;
case problem::base_tsp::CITIES:
fitness[i] = prob->objfun(my_pop[i]);
break;
}
}
decision_vector tmp_tour(Nv);
bool stop, changed;
size_t rnd_num, i2, pos1_c1, pos1_c2, pos2_c1, pos2_c2; //pos2_c1 denotes the position of city1 in parent2
fitness_vector fitness_tmp;
//InverOver main loop
for(int iter = 0; iter < m_gen; iter++) {
for(size_t i1 = 0; i1 < NP; i1++) {
tmp_tour = my_pop[i1];
pos1_c1 = unif_Nv();
stop = false;
changed = false;
while(!stop){
if(unif_01() < m_ri) {
rnd_num = unif_Nvless1();
pos1_c2 = (rnd_num == pos1_c1? Nv-1:rnd_num);
} else {
i2 = unif_NPless1();
i2 = (i2 == i1? NP-1:i2);
pos2_c1 = std::find(my_pop[i2].begin(),my_pop[i2].end(),tmp_tour[pos1_c1])-my_pop[i2].begin();
pos2_c2 = (pos2_c1 == Nv-1? 0:pos2_c1+1);
pos1_c2 = std::find(tmp_tour.begin(),tmp_tour.end(),my_pop[i2][pos2_c2])-tmp_tour.begin();
}
stop = (abs(pos1_c1-pos1_c2)==1 || static_cast<problem::base::size_type>(abs(pos1_c1-pos1_c2))==Nv-1);
if(!stop) {
changed = true;
if(pos1_c1<pos1_c2) {
for(size_t l=0; l < (double (pos1_c2-pos1_c1-1)/2); l++) {
std::swap(tmp_tour[pos1_c1+1+l],tmp_tour[pos1_c2-l]);
}
pos1_c1 = pos1_c2;
} else {
//inverts the section from c1 to c2 (see documentation Note3)
for(size_t l=0; l < (double (pos1_c1-pos1_c2-1)/2); l++) {
std::swap(tmp_tour[pos1_c2+l],tmp_tour[pos1_c1-l-1]);
}
pos1_c1 = (pos1_c2 == 0? Nv-1:pos1_c2-1);
}
}
} //end of while loop (looping over a single indvidual)
if(changed) {
switch(prob->get_encoding()) {
case problem::base_tsp::FULL:
fitness_tmp = prob->objfun(prob->full2cities(tmp_tour));
break;
case problem::base_tsp::RANDOMKEYS: //using "randomly" index 0 as a temporary template
fitness_tmp = prob->objfun(prob->cities2randomkeys(tmp_tour, pop.get_individual(0).cur_x));
break;
case problem::base_tsp::CITIES:
fitness_tmp = prob->objfun(tmp_tour);
break;
}
if(prob->compare_fitness(fitness_tmp,fitness[i1])) { //replace individual?
my_pop[i1] = tmp_tour;
fitness[i1][0] = fitness_tmp[0];
}
}
} // end of loop over population
} // end of loop over generations
//change representation of tour
for (size_t ii = 0; ii < NP; ii++) {
switch(prob->get_encoding()) {
case problem::base_tsp::FULL:
pop.set_x(ii,prob->cities2full(my_pop[ii]));
break;
case problem::base_tsp::RANDOMKEYS:
pop.set_x(ii,prob->cities2randomkeys(my_pop[ii],pop.get_individual(ii).cur_x));
break;
case problem::base_tsp::CITIES:
pop.set_x(ii,my_pop[ii]);
break;
}
}
} // end of evolve
// Selection implementation.
std::vector<std::pair<population::size_type,std::vector<population::individual_type>::size_type> >
worst_r_policy::select(const std::vector<population::individual_type> &immigrants, const population &dest) const
{
const population::size_type rate_limit = std::min<population::size_type>(get_n_individuals(dest),boost::numeric_cast<population::size_type>(immigrants.size()));
// Temporary vectors to store sorted indices of the populations.
std::vector<population::size_type> immigrants_idx(boost::numeric_cast<std::vector<population::size_type>::size_type>(immigrants.size()));
std::vector<population::size_type> dest_idx(boost::numeric_cast<std::vector<population::size_type>::size_type>(dest.size()));
// Fill in the arrays of indices.
iota(immigrants_idx.begin(),immigrants_idx.end(),population::size_type(0));
iota(dest_idx.begin(),dest_idx.end(),population::size_type(0));
// Sort the arrays of indices.
// From best to worst.
std::sort(immigrants_idx.begin(),immigrants_idx.end(),indirect_individual_sorter<std::vector<population::individual_type> >(immigrants,dest));
// From worst to best.
std::sort(dest_idx.begin(),dest_idx.end(),indirect_individual_sorter<population>(dest,dest));
std::reverse(dest_idx.begin(),dest_idx.end());
// Create the result.
std::vector<std::pair<population::size_type,std::vector<population::individual_type>::size_type> > result;
for (population::size_type i = 0; i < rate_limit; ++i) {
// Similar to fair policy, but replace unconditionally, without checking if the incoming individuals are better.
result.push_back(std::make_pair(dest_idx[i],immigrants_idx[i]));
}
return result;
}
void evaluate_population(population& p)
{
for(unsigned int i=0; i<p.size(); i++)
p[i].eval = evaluation(p[i].perm);
}
// Selection implementation.
std::vector<std::pair<population::size_type,std::vector<population::individual_type>::size_type> >
hv_fair_r_policy::select(const std::vector<population::individual_type> &immigrants, const population &dest) const
{
// Fall back to fair_r_policy when facing a single-objective problem.
if (dest.problem().get_f_dimension() == 1) {
return fair_r_policy(m_rate, m_type).select(immigrants, dest);
}
std::vector<population::individual_type> filtered_immigrants;
filtered_immigrants.reserve(immigrants.size());
// Keeps information on the original indexing of immigrants after we filter out the duplicates
std::vector<unsigned int> original_immigrant_indices;
original_immigrant_indices.reserve(immigrants.size());
// Remove the duplicates from the set of immigrants
std::vector<population::individual_type>::iterator im_it = (const_cast<std::vector<population::individual_type> &>(immigrants)).begin();
unsigned int im_idx = 0;
for( ; im_it != immigrants.end() ; ++im_it) {
decision_vector im_x((*im_it).cur_x);
bool equal = true;
for ( unsigned int idx = 0 ; idx < dest.size() ; ++idx ) {
decision_vector isl_x(dest.get_individual(idx).cur_x);
equal = true;
for (unsigned int d_idx = 0 ; d_idx < im_x.size() ; ++d_idx) {
if (im_x[d_idx] != isl_x[d_idx]) {
equal = false;
break;
}
}
if (equal) {
break;
}
}
if (!equal) {
filtered_immigrants.push_back(*im_it);
original_immigrant_indices.push_back(im_idx);
}
++im_idx;
}
// Computes the number of immigrants to be selected (accounting for the destination pop size)
const population::size_type rate_limit = std::min<population::size_type>(get_n_individuals(dest), boost::numeric_cast<population::size_type>(filtered_immigrants.size()));
// Defines the retvalue
std::vector<std::pair<population::size_type, std::vector<population::individual_type>::size_type> > result;
// Skip the remaining computation if there's nothing to do
if (rate_limit == 0) {
return result;
}
// Makes a copy of the destination population
population pop_copy(dest);
// Merge the immigrants to the copy of the destination population
for (population::size_type i = 0; i < rate_limit; ++i) {
pop_copy.push_back(filtered_immigrants[i].cur_x);
}
// Population fronts stored as indices of individuals.
std::vector< std::vector<population::size_type> > fronts_i = pop_copy.compute_pareto_fronts();
// Population fronts stored as fitness vectors of individuals.
std::vector< std::vector<fitness_vector> > fronts_f (fronts_i.size());
// Nadir point is established manually later, first point is a first "safe" candidate.
fitness_vector refpoint(pop_copy.get_individual(0).cur_f);
// Fill fronts_f with fitness vectors and establish the nadir point
for (unsigned int f_idx = 0 ; f_idx < fronts_i.size() ; ++f_idx) {
fronts_f[f_idx].resize(fronts_i[f_idx].size());
for (unsigned int p_idx = 0 ; p_idx < fronts_i[f_idx].size() ; ++p_idx) {
fronts_f[f_idx][p_idx] = fitness_vector(pop_copy.get_individual(fronts_i[f_idx][p_idx]).cur_f);
// Update the nadir point manually for efficiency.
for (unsigned int d_idx = 0 ; d_idx < fronts_f[f_idx][p_idx].size() ; ++d_idx) {
refpoint[d_idx] = std::max(refpoint[d_idx], fronts_f[f_idx][p_idx][d_idx]);
}
}
}
// Epsilon is added to nadir point
for (unsigned int d_idx = 0 ; d_idx < refpoint.size() ; ++d_idx) {
refpoint[d_idx] += m_nadir_eps;
}
// Vector for maintaining the original indices of points for augmented population as 0 and 1
std::vector<unsigned int> g_orig_indices(pop_copy.size(), 1);
unsigned int no_discarded_immigrants = 0;
// Store which front we process (start with the last front) and the number of processed individuals.
unsigned int front_idx = fronts_i.size(); // front_idx is equal to the size, since it's decremented right in the main loop
unsigned int processed_individuals = 0;
// Pairs of (islander index, islander exclusive hypervolume)
// Second item is updated later
std::vector<std::pair<unsigned int, double> > discarded_islanders;
std::vector<std::pair<unsigned int, double> > point_pairs;
// index of currently processed point in the point_pair vector.
// Initiated to its size (=0) in order to enforce the initial computation on penultimate front.
unsigned int current_point = point_pairs.size();
// Stops when we reduce the augmented population to the size of the original population or when the number of discarded islanders reaches the limit
while (processed_individuals < filtered_immigrants.size() && discarded_islanders.size() < rate_limit) {
// if current front was exhausted, load next one
if (current_point == point_pairs.size()) {
--front_idx;
// Compute contributions
std::vector<double> c;
// If there exist a dominated front for front at index front_idx
if (front_idx + 1 < fronts_f.size()) {
std::vector<fitness_vector> merged_front;
// Reserve the memory and copy the fronts
merged_front.reserve(fronts_f[front_idx].size() + fronts_f[front_idx + 1].size());
copy(fronts_f[front_idx].begin(), fronts_f[front_idx].end(), back_inserter(merged_front));
copy(fronts_f[front_idx + 1].begin(), fronts_f[front_idx +1].end(), back_inserter(merged_front));
hypervolume hv(merged_front, false);
c = hv.contributions(refpoint);
} else {
hypervolume hv(fronts_f[front_idx], false);
c = hv.contributions(refpoint);
}
// Initiate the pairs and sort by second item (exclusive volume)
point_pairs.resize(fronts_f[front_idx].size());
for(unsigned int i = 0 ; i < fronts_f[front_idx].size() ; ++i) {
point_pairs[i] = std::make_pair(i, c[i]);
}
current_point = 0;
std::sort(point_pairs.begin(), point_pairs.end(), sort_point_pairs_asc);
}
unsigned int orig_lc_idx = fronts_i[front_idx][point_pairs[current_point].first];
if (orig_lc_idx < dest.size()) {
discarded_islanders.push_back(std::make_pair(orig_lc_idx, 0.0));
} else {
++no_discarded_immigrants;
}
// Flag given individual as discarded
g_orig_indices[orig_lc_idx] = 0;
++processed_individuals;
++current_point;
}
// Number of non-discarded immigrants
unsigned int no_available_immigrants = boost::numeric_cast<unsigned int>(filtered_immigrants.size() - no_discarded_immigrants);
// Pairs of (immigrant index, immigrant exclusive hypervolume)
// Second item is updated later
std::vector<std::pair<unsigned int, double> > available_immigrants;
available_immigrants.reserve(no_available_immigrants);
for(unsigned int idx = dest.size() ; idx < pop_copy.size() ; ++idx) {
// If the immigrant was not discarded add it to the available set
if ( g_orig_indices[idx] == 1 ) {
available_immigrants.push_back(std::make_pair(idx, 0.0));
}
}
// Aggregate all points to establish the hypervolume contribution of available immigrants and discarded islanders
std::vector<fitness_vector> merged_fronts;
merged_fronts.reserve(pop_copy.size());
for(unsigned int idx = 0 ; idx < pop_copy.size() ; ++idx) {
merged_fronts.push_back(pop_copy.get_individual(idx).cur_f);
}
hypervolume hv(merged_fronts, false);
std::vector<std::pair<unsigned int, double> >::iterator it;
for(it = available_immigrants.begin() ; it != available_immigrants.end() ; ++it) {
(*it).second = hv.exclusive((*it).first, refpoint);
}
for(it = discarded_islanders.begin() ; it != discarded_islanders.end() ; ++it) {
(*it).second = hv.exclusive((*it).first, refpoint);
}
// Sort islanders and immigrants according to exclusive hypervolume
sort(available_immigrants.begin(), available_immigrants.end(), hv_fair_r_policy::ind_cmp);
sort(discarded_islanders.begin(), discarded_islanders.end(), hv_fair_r_policy::ind_cmp);
// Number of exchanges is the minimum of the number of non discarded immigrants and the number of discarded islanders
unsigned int no_exchanges = std::min(boost::numeric_cast<unsigned int>(available_immigrants.size()), boost::numeric_cast<unsigned int>(discarded_islanders.size()));
it = available_immigrants.begin();
std::vector<std::pair<unsigned int, double> >::reverse_iterator r_it = discarded_islanders.rbegin();
// Match the best immigrant (forward iterator) with the worst islander (reverse iterator) no_exchanges times.
for(unsigned int i = 0 ; i < no_exchanges ; ++i) {
// Break if any islander is better than an immigrant
if ((*r_it).second > (*it).second) {
break;
}
// Push the pair (islander_idx, fixed_immigrant_idx) to the results
result.push_back(std::make_pair((*r_it).first, original_immigrant_indices[(*it).first - dest.size()]));
++r_it;
++it;
}
return result;
}
// Evolve method.
void base_nlopt::evolve(population &pop) const
{
// Useful variables.
const problem::base &problem = pop.problem();
if (problem.get_f_dimension() != 1) {
pagmo_throw(value_error,"this algorithm does not support multi-objective optimisation");
}
const problem::base::c_size_type c_size = problem.get_c_dimension();
const problem::base::c_size_type ec_size = problem.get_c_dimension() - problem.get_ic_dimension();
if (c_size && !m_constrained) {
pagmo_throw(value_error,"this algorithm does not support constraints");
}
if (ec_size && m_only_ineq) {
pagmo_throw(value_error,"this algorithm does not support equality constraints");
}
const problem::base::size_type cont_size = problem.get_dimension() - problem.get_i_dimension();
if (!cont_size) {
pagmo_throw(value_error,"the problem has no continuous part");
}
// Do nothing if the population is empty.
if (!pop.size()) {
return;
}
// Extract the best individual and set the inital point
const population::size_type best_ind_idx = pop.get_best_idx();
const population::individual_type &best_ind = pop.get_individual(best_ind_idx);
// Structure to pass data to the objective function wrapper.
nlopt_wrapper_data data_objfun;
data_objfun.prob = &problem;
data_objfun.x.resize(problem.get_dimension());
data_objfun.dx.resize(problem.get_dimension());
data_objfun.f.resize(1);
// Structure to pass data to the constraint function wrapper.
std::vector<nlopt_wrapper_data> data_constrfun(boost::numeric_cast<std::vector<nlopt_wrapper_data>::size_type>(c_size));
for (problem::base::c_size_type i = 0; i < c_size; ++i) {
data_constrfun[i].prob = &problem;
data_constrfun[i].x.resize(problem.get_dimension());
data_constrfun[i].dx.resize(problem.get_dimension());
data_constrfun[i].c.resize(problem.get_c_dimension());
data_constrfun[i].c_comp = i;
}
// Main NLopt call.
nlopt::opt opt(m_algo, problem.get_dimension());
m_opt = opt;
// Sets local optimizer for aug_lag methods, do nothing otherwise
set_local(problem.get_dimension());
m_opt.set_lower_bounds(problem.get_lb());
m_opt.set_upper_bounds(problem.get_ub());
m_opt.set_min_objective(objfun_wrapper, &data_objfun);
for (problem::base::c_size_type i =0; i<ec_size; ++i) {
m_opt.add_equality_constraint(constraints_wrapper, &data_constrfun[i], problem.get_c_tol().at(i));
}
for (problem::base::c_size_type i =ec_size; i<c_size; ++i) {
m_opt.add_inequality_constraint(constraints_wrapper, &data_constrfun[i], problem.get_c_tol().at(i));
}
m_opt.set_ftol_abs(m_ftol);
m_opt.set_xtol_abs(m_xtol);
m_opt.set_maxeval(m_max_iter);
//nlopt::result result;
double dummy;
decision_vector x0(best_ind.cur_x);
m_opt.optimize(x0, dummy);
pop.set_x(best_ind_idx,x0);
}