def nsga2_replacement(new_pop, old_pop):
"""
Replaces the old population with the new population using NSGA-II
replacement. Both new and old populations are combined, pareto fronts
and crowding distance are calculated, and the replacement population is
computed based on crowding distance per pareto front.
:param new_pop: The new population (e.g. after selection, variation, &
evaluation).
:param old_pop: The previous generation population.
:return: The 'POPULATION_SIZE' new population.
"""
# Combine both populations (R_t = P_t union Q_t)
new_pop.extend(old_pop)
# Compute the pareto fronts and crowding distance
pareto = compute_pareto_metrics(new_pop)
# Size of the new population
pop_size = params['POPULATION_SIZE']
# New population to replace the last one
temp_pop, i = [], 0
while len(temp_pop) < pop_size:
# Populate the replacement population
if len(pareto.fronts[i]) <= pop_size - len(temp_pop):
temp_pop.extend(pareto.fronts[i])
else:
# Sort the current pareto front with respect to crowding distance.
pareto.fronts[i] = sorted(
pareto.fronts[i],
key=lambda item: pareto.crowding_distance[item])
# Get number of individuals to add in temp to achieve the pop_size
diff_size = pop_size - len(temp_pop)
# Extend the replacement population
temp_pop.extend(pareto.fronts[i][:diff_size])
# Increment counter.
i += 1
return temp_pop
def nsga2_replacement(new_pop, old_pop):
"""
Replaces the old population with the new population using NSGA-II
replacement. Both new and old populations are combined, pareto fronts
and crowding distance are calculated, and the replacement population is
computed based on crowding distance per pareto front.
:param new_pop: The new population (e.g. after selection, variation, &
evaluation).
:param old_pop: The previous generation population.
:return: The 'POPULATION_SIZE' new population.
"""
# Combine both populations (R_t = P_t union Q_t)
new_pop.extend(old_pop)
# Compute the pareto fronts and crowding distance
pareto = compute_pareto_metrics(new_pop)
# Size of the new population
pop_size = params['POPULATION_SIZE']
# New population to replace the last one
temp_pop, i = [], 0
while len(temp_pop) < pop_size:
# Populate the replacement population
if len(pareto.fronts[i]) <= pop_size - len(temp_pop):
temp_pop.extend(pareto.fronts[i])
else:
# Sort the current pareto front with respect to crowding distance.
pareto.fronts[i] = sorted(pareto.fronts[i],
key=lambda item:
pareto.crowding_distance[item])
# Get number of individuals to add in temp to achieve the pop_size
diff_size = pop_size - len(temp_pop)
# Extend the replacement population
temp_pop.extend(pareto.fronts[i][:diff_size])
# Increment counter.
i += 1
return temp_pop
def nsga2_selection(population):
"""Apply NSGA-II selection operator on the *population*. Usually, the
size of *population* will be larger than *k* because any individual
present in *population* will appear in the returned list at most once.
Having the size of *population* equals to *k* will have no effect other
than sorting the population according to their front rank. The
list returned contains references to the input *population*. For more
details on the NSGA-II operator see [Deb2002]_.
:param population: A population from which to select individuals.
:returns: A list of selected individuals.
.. [Deb2002] Deb, Pratab, Agarwal, and Meyarivan, "A fast elitist
non-dominated sorting genetic algorithm for multi-objective
optimization: NSGA-II", 2002.
"""
selection_size = params['GENERATION_SIZE']
tournament_size = params['TOURNAMENT_SIZE']
# Initialise list of tournament winners.
winners = []
# The flag "INVALID_SELECTION" allows for selection of invalid individuals.
if params['INVALID_SELECTION']:
available = population
else:
available = [i for i in population if not i.invalid]
# Compute pareto front metrics.
pareto = compute_pareto_metrics(available)
while len(winners) < selection_size:
# Return the single best competitor.
winners.append(pareto_tournament(available, pareto, tournament_size))
return winners
def nsga2_selection(population):
"""Apply NSGA-II selection operator on the *population*. Usually, the
size of *population* will be larger than *k* because any individual
present in *population* will appear in the returned list at most once.
Having the size of *population* equals to *k* will have no effect other
than sorting the population according to their front rank. The
list returned contains references to the input *population*. For more
details on the NSGA-II operator see [Deb2002]_.
:param population: A population from which to select individuals.
:returns: A list of selected individuals.
.. [Deb2002] Deb, Pratab, Agarwal, and Meyarivan, "A fast elitist
non-dominated sorting genetic algorithm for multi-objective
optimization: NSGA-II", 2002.
"""
selection_size = params['GENERATION_SIZE']
tournament_size = params['TOURNAMENT_SIZE']
# Initialise list of tournament winners.
winners = []
# The flag "INVALID_SELECTION" allows for selection of invalid individuals.
if params['INVALID_SELECTION']:
available = population
else:
available = [i for i in population if not i.invalid]
# Compute pareto front metrics.
pareto = compute_pareto_metrics(available)
while len(winners) < selection_size:
# Return the single best competitor.
winners.append(pareto_tournament(available, pareto, tournament_size))
return winners
def get_moo_stats(individuals, end):
"""
Generate the statistics for an evolutionary run with multiple objectives.
Save statistics to utilities.trackers.stats_list. Print statistics. Save
fitness plot information.
:param individuals: A population of individuals for which to generate
statistics.
:param end: Boolean flag for indicating the end of an evolutionary run.
:return: Nothing.
"""
# Compute the pareto front metrics for the population.
pareto = compute_pareto_metrics(individuals)
# Save first front in trackers. Sort arbitrarily along first objective.
trackers.best_ever = sorted(pareto.fronts[0], key=lambda x: x.fitness[0])
# Store stats about pareto fronts.
stats['pareto_fronts'] = len(pareto.fronts)
stats['first_front'] = len(pareto.fronts[0])
if end or params['VERBOSE'] or not params['DEBUG']:
# Update all stats.
update_stats(individuals, end)
# Save fitness plot information
if params['SAVE_PLOTS'] and not params['DEBUG']:
# Initialise empty array for fitnesses for all inds on first pareto
# front.
all_arr = [[] for _ in range(params['FITNESS_FUNCTION'].num_obj)]
# Generate array of fitness values.
fitness_array = [ind.fitness for ind in trackers.best_ever]
# Add paired fitnesses to array for graphing.
for fit in fitness_array:
for o in range(params['FITNESS_FUNCTION'].num_obj):
all_arr[o].append(fit[o])
if not end:
trackers.first_pareto_list.append(all_arr)
# Append empty array to best fitness list.
trackers.best_fitness_list.append([])
# Get best fitness for each objective.
for o, ff in \
enumerate(params['FITNESS_FUNCTION'].fitness_functions):
# Get sorted list of all fitness values for objective "o"
fits = sorted(all_arr[o], reverse=ff.maximise)
# Append best fitness to trackers list.
trackers.best_fitness_list[-1].append(fits[0])
if params['VERBOSE'] or end:
# Plot best fitness for each objective.
for o, ff in \
enumerate(params['FITNESS_FUNCTION'].fitness_functions):
to_plot = [i[o] for i in trackers.best_fitness_list]
# Plot fitness data for objective o.
plotname = ff.__class__.__name__ + str(o)
save_plot_from_data(to_plot, plotname)
# TODO: PonyGE2 can currently only plot moo problems with 2 objectives.
# Check that the number of fitness objectives is not greater than 2
if params['FITNESS_FUNCTION'].num_obj > 2:
s = "stats.stats.get_moo_stats\n" \
"Warning: Plotting of more than 2 simultaneous " \
"objectives is not yet enabled in PonyGE2."
print(s)
else:
save_pareto_fitness_plot()
# Print statistics
if params['VERBOSE'] and not end:
print_generation_stats()
print_first_front_stats()
elif not params['SILENT']:
# Print simple display output.
perc = stats['gen'] / (params['GENERATIONS'] + 1) * 100
stdout.write("Evolution: %d%% complete\r" % perc)
stdout.flush()
# Generate test fitness on regression problems
if hasattr(params['FITNESS_FUNCTION'], "training_test") and end:
for ind in trackers.best_ever:
# Iterate over all individuals in the first front.
# Save training fitness.
ind.training_fitness = copy(ind.fitness)
# Evaluate test fitness.
ind.test_fitness = params['FITNESS_FUNCTION'](ind, dist='test')
# Set main fitness as training fitness.
ind.fitness = ind.training_fitness
# Save stats to list.
if params['VERBOSE'] or (not params['DEBUG'] and not end):
trackers.stats_list.append(copy(stats))
# Save stats to file.
if not params['DEBUG']:
if stats['gen'] == 0:
save_stats_headers(stats)
save_stats_to_file(stats, end)
if params['SAVE_ALL']:
save_first_front_to_file(stats, end, stats['gen'])
elif params['VERBOSE'] or end:
save_first_front_to_file(stats, end)
if end and not params['SILENT']:
print_final_moo_stats()
def get_moo_stats(individuals, end):
"""
Generate the statistics for an evolutionary run with multiple objectives.
Save statistics to utilities.trackers.stats_list. Print statistics. Save
fitness plot information.
:param individuals: A population of individuals for which to generate
statistics.
:param end: Boolean flag for indicating the end of an evolutionary run.
:return: Nothing.
"""
# Compute the pareto front metrics for the population.
pareto = compute_pareto_metrics(individuals)
# Save first front in trackers. Sort arbitrarily along first objective.
trackers.best_ever = sorted(pareto.fronts[0], key=lambda x: x.fitness[0])
# Store stats about pareto fronts.
stats['pareto_fronts'] = len(pareto.fronts)
stats['first_front'] = len(pareto.fronts[0])
if end or params['VERBOSE'] or not params['DEBUG']:
# Update all stats.
update_stats(individuals, end)
# Save fitness plot information
if params['SAVE_PLOTS'] and not params['DEBUG']:
# Initialise empty array for fitnesses for all inds on first pareto
# front.
all_arr = [[] for _ in range(params['FITNESS_FUNCTION'].num_obj)]
# Generate array of fitness values.
fitness_array = [ind.fitness for ind in trackers.best_ever]
# Add paired fitnesses to array for graphing.
for fit in fitness_array:
for o in range(params['FITNESS_FUNCTION'].num_obj):
all_arr[o].append(fit[o])
if not end:
trackers.first_pareto_list.append(all_arr)
# Append empty array to best fitness list.
trackers.best_fitness_list.append([])
# Get best fitness for each objective.
for o, ff in \
enumerate(params['FITNESS_FUNCTION'].fitness_functions):
# Get sorted list of all fitness values for objective "o"
fits = sorted(all_arr[o], reverse=ff.maximise)
# Append best fitness to trackers list.
trackers.best_fitness_list[-1].append(fits[0])
if params['VERBOSE'] or end:
# Plot best fitness for each objective.
for o, ff in \
enumerate(params['FITNESS_FUNCTION'].fitness_functions):
to_plot = [i[o] for i in trackers.best_fitness_list]
# Plot fitness data for objective o.
plotname = ff.__class__.__name__ + str(o)
save_plot_from_data(to_plot, plotname)
# TODO: PonyGE2 can currently only plot moo problems with 2 objectives.
# Check that the number of fitness objectives is not greater than 2
if params['FITNESS_FUNCTION'].num_obj > 2:
s = "stats.stats.get_moo_stats\n" \
"Warning: Plotting of more than 2 simultaneous " \
"objectives is not yet enabled in PonyGE2."
print(s)
else:
save_pareto_fitness_plot()
# Print statistics
if params['VERBOSE'] and not end:
print_generation_stats()
print_first_front_stats()
elif not params['SILENT']:
# Print simple display output.
perc = stats['gen'] / (params['GENERATIONS'] + 1) * 100
stdout.write("Evolution: %d%% complete\r" % perc)
stdout.flush()
# Generate test fitness on regression problems
if hasattr(params['FITNESS_FUNCTION'], "training_test") and end:
for ind in trackers.best_ever:
# Iterate over all individuals in the first front.
# Save training fitness.
ind.training_fitness = copy(ind.fitness)
# Evaluate test fitness.
ind.test_fitness = params['FITNESS_FUNCTION'](ind, dist='test')
# Set main fitness as training fitness.
ind.fitness = ind.training_fitness
# Save stats to list.
if params['VERBOSE'] or (not params['DEBUG'] and not end):
trackers.stats_list.append(copy(stats))
# Save stats to file.
if not params['DEBUG']:
if stats['gen'] == 0:
save_stats_headers(stats)
save_stats_to_file(stats, end)
if params['SAVE_ALL']:
save_first_front_to_file(stats, end, stats['gen'])
elif params['VERBOSE'] or end:
save_first_front_to_file(stats, end)
if end and not params['SILENT']:
print_final_moo_stats()