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()
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
class GeneticAlgorithm(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,
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.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")
#instantiating population
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.generation = 0
self.population_stats = []
self.current_stats = None
def initialize(self):
"""
Initializes population, cache and storage
"""
self.population.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)
"""
self.population.fit_individuals(self.fitness_evaluator, self.cache, eval_callback=self.store_individual)
stats = self.stat_population(**options)
self.population.scale_individuals(stats)
def stat_population(self, **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.
"""
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
self.population_stats.append(stats)
self.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.population_stats[self.generation][top_key] == self.population_stats[self.generation-1][top_key]:
stats['idle_cycles'] = self.population_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 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
"""
new_population = self.population.evolve(**options)
self.population = new_population
self.generation = new_population.generation_number
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_terminate():
break
self.evolve_population(global_stats=self.population_stats, last_stats=self.current_stats, ga_engine=self)
self.evaluate_population(**options)
return self.best_individual()
