def test_best_indivividual(self):
protogene_a = ProtoGene(IntGene, 'a')
protogenome = ProtoGenome([protogene_a])
pop = Population(protogenome, optimization_mode='min')
pop.initialize()
for i in range(pop.size):
pop.individuals[i].score = i
best_ind = pop.best_individual()
assert pop.individuals[0].score == best_ind.score == 0
assert pop.individuals[0].get_hash() == best_ind.get_hash()
def test_sort(self):
protogene_a = ProtoGene(IntGene, 'a')
protogenome = ProtoGenome([protogene_a])
pop = Population(protogenome, optimization_mode='min')
pop.initialize()
for i in range(pop.size):
pop.individuals[i].score = i
pop.sort()
assert pop.individuals[0].score == 0
pop_2 = Population(protogenome, optimization_mode='max')
pop_2.initialize(individuals=pop.individuals)
pop_2.sort()
assert pop_2.individuals[-1].score == 0
assert pop_2.individuals[-1].score == pop.individuals[0].score
def test_fit_individuals(self):
protogene_a = ProtoGene(IntGene, 'a', value=1)
protogenome = ProtoGenome([protogene_a])
pop = Population(protogenome, optimization_mode='min')
pop.initialize()
def fitness_evaluator(g):
return g.get_value('a')
pop.fit_individuals(fitness_evaluator)
for individual in pop.individuals:
assert individual.score == 1
dict_cache = DictCache()
pop.fit_individuals(fitness_evaluator, cache=dict_cache)
for individual in pop.individuals:
assert individual.score == 1
self.total_score = 0
def eval_callback(hash, individual):
self.total_score += individual.score
pop.fit_individuals(fitness_evaluator, cache=dict_cache, eval_callback=eval_callback)
assert self.total_score == pop.size
def test_evolve(self):
protogene_a = ProtoGene(IntGene, 'a', value=1)
protogenome = ProtoGenome([protogene_a])
pop = Population(protogenome, optimization_mode='min')
pop.initialize()
def fitness_evaluator(g):
return g.get_value('a')
pop.fit_individuals(fitness_evaluator)
new_population = pop.evolve()
class MultiGeneticAlgorithm(object):
"""
Genetic algorithm class
:param protogenome: instance of :class:`genepi.core.protogenome.Protogenome`
:param fitness_evaluator: the fitness function
:param termination_criteria: one or more termination criteria for evolution
:param termination_criteria_options: options for each termination criteria, each option
set is passed as a dictionary. Options must be passed in the same order as termination
criteria
:param population_size: size of the population
"""
def __init__(self,
protogenome,
fitness_evaluator,
optimization_mode='min',
termination_criteria=stopcriteria.convergence_stop,
termination_criteria_options={},
population_size=POPULATION_DEFAULT_SIZE,
num_populations=4,
isolated_cycles=3,
step_callback=None,
selection_method=None,
elitism=True,
num_parents=2,
crossover_method=None,
crossover_probability=0.1,
crossover_wrapper_method = None,
mutation_wrapper_method=None,
storage_instance=None,
cache_instance=None,
**options):
"""
"""
self.protogenome = protogenome
self.fitness_evaluator = fitness_evaluator
self.optimization_mode = optimization_mode
self.termination_criteria = termination_criteria
self.termination_criteria_options = termination_criteria_options
self.population_size = population_size
self.num_populations = num_populations
self.isolated_cycles = isolated_cycles
self.selection_method = selection_method
self.elitism = elitism
self.num_parents = num_parents
self.step_callback = step_callback
#default crossover and selections are handled by population
self.crossover_method = crossover_method
self.crossover_probability = crossover_probability
#mutation wrapper
self.mutation_wrapper_method = mutation_wrapper_method
#crossover wrapper
self.crossover_wrapper_method = crossover_wrapper_method
#cache configuration
if cache_instance is None:
self.cache = DictCache()
else:
self.cache = cache_instance
#storage configuration
if storage_instance:
self.storage = storage_instance
else:
self.storage = None
self.options = options
#termination criteria and options check
if type(self.termination_criteria) == type(list):
if type(self.termination_criteria_options) != type(list):
raise ValueError("You must pass options for each termination criteria")
if len(self.termination_criteria_options) != len(self.termination_criteria):
raise ValueError("You must pass options for each termination criteria")
self.populations= [];
self.population = Population(self.protogenome,
size=self.population_size,
optimization_mode=self.optimization_mode,
selection_method=self.selection_method,
crossover_method=self.crossover_method,
crossover_probability=self.crossover_probability,
elitism=self.elitism,
num_parents=self.num_parents,
mutation_wrapper_method=self.mutation_wrapper_method,
crossover_wrapper_method=self.crossover_wrapper_method)
self.population_stats = {}
self.current_stats = {}
for i in range(self.num_populations):
#instantiating population
pop = Population(self.protogenome,
size=self.population_size,
optimization_mode=self.optimization_mode,
selection_method=self.selection_method,
crossover_method=self.crossover_method,
crossover_probability=self.crossover_probability,
elitism=self.elitism,
num_parents=self.num_parents,
mutation_wrapper_method=self.mutation_wrapper_method,
crossover_wrapper_method=self.crossover_wrapper_method)
self.populations.append(pop)
self.population_stats[i] = {}
self.current_stats[i] = None
self.generation = 0
self.main_stats= {}
self.current_main_stats = None
def initialize(self):
"""
Initializes population, cache and storage
"""
for i in range(self.num_populations):
self.populations[i].initialize()
self.cache.initialize()
if self.storage:
self.storage.initialize()
def should_terminate(self):
"""
Called after each evolution cycle
"""
if type(self.termination_criteria) == type(list()):
for i,criterium in enumerate(self.termination_criteria):
if criterium(self, **self.termination_criteria_options[i]):
return True
else:
if self.termination_criteria(self, **self.termination_criteria_options):
return True
return False
def store_individual(self, hash, individual):
"""
Store an individual in the storage backend.
"""
if self.storage:
self.storage.write_individual(hash, self.generation, individual )
def evaluate_population(self, **options):
"""
Evaluate current population, the following operations are performed:
* run fitness_evaluator (compute score on each individual)
* calculate statistics
* scale population individuals (compute scaled_score on each individual)
"""
for i in range(self.num_populations):
self.populations[i].fit_individuals(self.fitness_evaluator, self.cache, eval_callback=self.store_individual)
stats = self.stat_population(pop_id=i, **options)
self.populations[i].scale_individuals(stats)
def stat_population(self, pop_id=None,**options):
"""
Compute statistics for current population: min max and average scores, idle cycles.
If a storage is set, stats are written to storage backend.
After computation, stats are stored in population_stats property, that contains stats
for all generations.
the current_generation property is set to stats.
"""
if pop_id is not None:
if pop_id not in self.population_stats:
self.population_stats[pop_id] = {}
scores = [individual.score for individual in self.populations[pop_id].individuals]
else:
scores = [individual.score for individual in self.population.individuals]
avg_score = sum(scores) / len(scores)
min_score = min(scores)
max_score = max(scores)
stats = { 'avg_score' : avg_score, 'min_score' : min_score, 'max_score' : max_score,
'generation' : self.generation }
#setting stats properties
if pop_id is not None:
self.population_stats[pop_id][self.generation] = stats
self.current_stats[pop_id] = stats
if options.get('debug', None):
print "pop_id:", pop_id, stats
else:
self.main_stats[self.generation] = stats
self.main_current_stats = stats
#idle cycles calculation: it counts consequent generations without improvements
if self.optimization_mode == 'max':
top_key = 'max_score'
else:
top_key = 'min_score'
if self.generation == 0:
stats['idle_cycles'] = 0
else:
if self.main_stats[self.generation][top_key] == self.main_stats[self.generation-1][top_key]:
stats['idle_cycles'] = self.main_stats[self.generation-1]['idle_cycles'] + 1
else:
stats['idle_cycles'] = 0
if options.get('debug', None):
print stats
if self.storage:
self.storage.write_population_stats(self.generation, stats)
return stats
def should_merge_populations(self):
"""
"""
if self.populations[0].generation_number % self.isolated_cycles == 0:
return True
return False
def merge_populations(self, **options):
"""
"""
all_individuals = []
for i in range(self.num_populations):
all_individuals += self.populations[i].individuals
all_individuals.sort(self.population.cmp_individual)
new_individuals = all_individuals[:self.population_size]
new_population = self.population.copy(individuals=new_individuals)
self.population = new_population
stats = self.stat_population( **options)
self.population.scale_individuals(stats)
self.generation += 1
self.mix_populations(all_individuals, new_individuals)
def mix_populations(self, all_individuals, champions):
"""
"""
for i in range(self.num_populations):
packet = []
"""
for d in range(10):
packet.append(random.choice(champions))
"""
for x in range(self.population_size):
item = (random.choice(all_individuals))
packet.append(item)
all_individuals.remove(item)
self.populations[i].initialize(individuals=packet)
self.populations[i].sort()
def best_individual(self):
"""
Returns best individual in population (relies on Population method)
"""
return self.population.best_individual()
def evolve_population(self, **options):
"""
creates a new population with population.evolve, sets the new population as
the current and increment generation.
After this the population in the previous generation is lost
"""
for i in range(self.num_populations):
new_population = self.populations[i].evolve(**options)
self.populations[i] = new_population
def evolve(self, **options):
"""
Performs the evolution cycle.
This is the main method that should be normally called.
Evolution goes on until a termination criterium becomes True.
At the end the best individual is returned.
"""
if not self.termination_criteria:
raise TypeError("You Must set one or more termination criteria")
self.initialize()
self.evaluate_population(**options)
while 1:
if self.should_merge_populations():
self.merge_populations(**options)
if self.should_terminate():
break
self.evolve_population(global_stats=self.population_stats, last_stats=self.current_stats, ga_engine=self)
self.evaluate_population(**options)
#self.merge_populations(**options)
return self.best_individual()
def __init__(self,
protogenome,
fitness_evaluator,
optimization_mode='min',
termination_criteria=stopcriteria.convergence_stop,
termination_criteria_options={},
population_size=POPULATION_DEFAULT_SIZE,
num_populations=4,
isolated_cycles=3,
step_callback=None,
selection_method=None,
elitism=True,
num_parents=2,
crossover_method=None,
crossover_probability=0.1,
crossover_wrapper_method = None,
mutation_wrapper_method=None,
storage_instance=None,
cache_instance=None,
**options):
"""
"""
self.protogenome = protogenome
self.fitness_evaluator = fitness_evaluator
self.optimization_mode = optimization_mode
self.termination_criteria = termination_criteria
self.termination_criteria_options = termination_criteria_options
self.population_size = population_size
self.num_populations = num_populations
self.isolated_cycles = isolated_cycles
self.selection_method = selection_method
self.elitism = elitism
self.num_parents = num_parents
self.step_callback = step_callback
#default crossover and selections are handled by population
self.crossover_method = crossover_method
self.crossover_probability = crossover_probability
#mutation wrapper
self.mutation_wrapper_method = mutation_wrapper_method
#crossover wrapper
self.crossover_wrapper_method = crossover_wrapper_method
#cache configuration
if cache_instance is None:
self.cache = DictCache()
else:
self.cache = cache_instance
#storage configuration
if storage_instance:
self.storage = storage_instance
else:
self.storage = None
self.options = options
#termination criteria and options check
if type(self.termination_criteria) == type(list):
if type(self.termination_criteria_options) != type(list):
raise ValueError("You must pass options for each termination criteria")
if len(self.termination_criteria_options) != len(self.termination_criteria):
raise ValueError("You must pass options for each termination criteria")
self.populations= [];
self.population = Population(self.protogenome,
size=self.population_size,
optimization_mode=self.optimization_mode,
selection_method=self.selection_method,
crossover_method=self.crossover_method,
crossover_probability=self.crossover_probability,
elitism=self.elitism,
num_parents=self.num_parents,
mutation_wrapper_method=self.mutation_wrapper_method,
crossover_wrapper_method=self.crossover_wrapper_method)
self.population_stats = {}
self.current_stats = {}
for i in range(self.num_populations):
#instantiating population
pop = Population(self.protogenome,
size=self.population_size,
optimization_mode=self.optimization_mode,
selection_method=self.selection_method,
crossover_method=self.crossover_method,
crossover_probability=self.crossover_probability,
elitism=self.elitism,
num_parents=self.num_parents,
mutation_wrapper_method=self.mutation_wrapper_method,
crossover_wrapper_method=self.crossover_wrapper_method)
self.populations.append(pop)
self.population_stats[i] = {}
self.current_stats[i] = None
self.generation = 0
self.main_stats= {}
self.current_main_stats = None
def test_mutate(self):
pop = Population(self.protogenome)
pop.initialize()
pop.mutate()
def test_copy(self):
pop = Population(self.protogenome)
pop.initialize()
pop2 = pop.copy()
def test_initialize(self):
pop = Population(self.protogenome)
pop.initialize()
assert len(pop.individuals) == pop.size
