void running_stat_add_sample( running_statistics* data, float new_value ) {
/* oldest value is the current index as it's one past the last value added */
float old_value = data->arr[data->idx];
/* Remove oldest value from sum if the index has wrapped around */
if( data->wrapped ) {
/* remove old value from sum using Kahan summation */
data->sum = kahan_sum( &data->kahan_sub, -old_value, data->sum );
if( data->cmpt_var ) {
variance_remove( &data->var, old_value );
}
}
/* Add new value in place of oldest value */
data->arr[data->idx++] = new_value;
/* Add new value to sum using Kahan summation */
data->sum = kahan_sum( &data->kahan_add, new_value, data->sum );
/* Reset idx to 0 when it reaches max_sz and remember that it wrapped */
if( data->idx >= data->max_sz ) {
data->idx = 0;
data->wrapped = 1;
}
if( data->cmpt_var ) {
variance_add( &data->var, new_value );
}
}
utils::PopulationStatistics utils::PopulationStatistics::compute(std::vector<FitnessPair>& fitnesses, std::shared_ptr<FitnessCore>& fitness_core) {
utils::PopulationStatistics::global_index++;
utils::PopulationStatistics result = utils::PopulationStatistics();
// median and quartile calculation depends on the data being sorted
std::sort(fitnesses.begin(), fitnesses.end(), [](FitnessPair& a, FitnessPair&b) { return a.fitness < b.fitness; });
unsigned int num_elements = (unsigned int) fitnesses.size();
result.min = fitnesses.front().fitness;
result.max = fitnesses.back().fitness;
result.mean = kahan_sum(fitnesses) / num_elements;
result.median = num_elements % 2 == 0 ? ((fitnesses[num_elements / 2].fitness + fitnesses[num_elements / 2 + 1].fitness) / 2.0) : (fitnesses[num_elements / 2].fitness);
result.lower_quartile = fitnesses[num_elements / 4].fitness;
result.upper_quartile = fitnesses[(3 * num_elements) / 4].fitness;
return result;
}
