void lbann_callback_summary::on_epoch_end(model *m) {
prof_region_begin("summary-epoch", prof_colors[0], false);
for (const auto& met : m->get_metrics()) {
EvalType train_score = met->get_mean_value(m->get_execution_mode());
// Replace spaces with _ for consistency.
std::string metric_name = met->name();
std::transform(metric_name.begin(), metric_name.end(), metric_name.begin(),
[] (char c) { return c == ' ' ? '_' : c; });
std::string phase = "train_" + metric_name;
m_summarizer->reduce_scalar(phase, train_score, m->get_cur_step());
}
save_histograms(m);
m_summarizer->flush();
prof_region_end("summary-epoch", false);
}
/* end of tasks */
int event_exit(){
/* show block number and the analyzing ratio */
fprintf(stderr,"\n");
show_blk_num(1);
save_scalers();
/* DST exit tasks */
dst_exit();
if(rootflag == 1)
root_exit();
save_histograms();
hb_exit(shmflag);
return 0;
}
void GA::step()
{
pop_vector children;
children.clear();
children.reserve(m_population_size);
// elitism
unsigned int elite_size = m_population_size * m_elitism_rate;
for (unsigned int i=0; i<elite_size; i++) {
children.push_back(m_population[i]->copy());
}
// crossover
while (children.size() < (m_population_size - elite_size)) {
// select parents
int p1 = roulette();
int p2 = roulette();
// crossover parents
crossover(m_population[p1], m_population[p2], children);
}
// mutation
mutate(children);
// update population and save last population
update(children);
// evaluate fitness and sort
evaluate_fitness();
// log, save generation
log();
save_histograms();
}
void lbann_callback_summary::on_train_begin(model *m) {
save_histograms(m);
}