def initialize(self):
super().initialize()
# Generate a new population
self.population = Population(
size=self.population_size,
spawning_pool=self.spawning_pool,
)
for individual in self.population:
individual.fitness_method = self.fitness
# Clear the best individuals historical cache
self.best_individuals.clear()
class SimpleGA(GeneticAlgorithm):
""" Simple implementation of a GeneticAlgorithm
This subclass implements the basic behavior of a genetic algorithm with some
degree of configuration.
"""
def __init__(
self,
stop_condition,
population_size,
spawning_pool,
fitness,
selection,
recombination,
replacement,
mutation=None,
diversity=None,
p_recombination=0.9,
p_mutation=0.1,
replacement_rate=1.0,
):
""" Initializes this instance.
:param stop_condition: The condition to be met in order to stop the
genetic algorithm.
:param population_size: The size this population should have.
:param spawning_pool: The object that generates individuals.
:param fitness: The method to evaluate individuals. It's expected to be
a callable that returns a float value where the higher the value,
the better the individual. Instances of subclasses of class Fitness
can be used for this purpose.
:param selection: The method to select individuals of the population to
recombine.
:param replacement: The method that will add and remove individuals from
the population given the set of old individuals (i.e. the ones on
the population before the evolution step) and new individuals (i.e.
the offspring).
:param recombination: The method to recombine parents in order to
generate an offspring with characteristics of the parents. If none,
no recombination will be applied.
:param mutation: The method to mutate an individual. If none, no
mutation over the individual will be applied. If not provided, no
mutation is performed.
:param diversity: The method to compute the diversity of a sequence of
individuals generated by this SpawningPool instance. Is expected to
be a function that generates a diversity representation given a
subset of individuals. Instances of subclasses of class Diversity
can be used for this purpose.
:param p_recombination: The odds for recombination method to be
performed over a set of selected individuals to generate progeny. If
not performed, progeny will be the parents. Must be a value between
0 and 1 (both included). If not provided, defaults to 1.0.
:param p_mutation: The odds for mutation method to be performed over a
progeny. It's applied once for each individual. If not performed the
individuals will not be modified. Must be a value between 0 and 1
(both included). If not provided, it defaults to 0.0 (no mutation is
performed).
:param replacement_rate: The rate of individuals to be replaced in each
step of the algorithm. Must be a float value in the (0, 1] interval.
:raises WrongValueForIntervalError: If any of the bounded values fall
out of their respective intervals.
:raises NotAProbabilityError: If a value was expected to be a
probability and it wasn't.
:raises UnexpectedClassError: If any of the input variables doesn't
follow the contract required (i.e. doesn't inherit from a predefined
class).
"""
super().__init__(stop_condition=stop_condition)
self.population_size = population_size
self.spawning_pool = spawning_pool
self.fitness = fitness
self.offspring_size = int(math.ceil(population_size * replacement_rate))
self.selection = selection
self.recombination = recombination
self.mutation = mutation or NoMutation()
self.diversity = diversity
self.replacement = replacement
self.replacement_rate = replacement_rate
self.p_recombination = p_recombination
self.p_mutation = p_mutation
self.selection_size = len(
inspect.signature(recombination.__call__).parameters
)
self.population = None
self.best_individuals = []
def initialize(self):
super().initialize()
# Generate a new population
self.population = Population(
size=self.population_size,
spawning_pool=self.spawning_pool,
)
for individual in self.population:
individual.fitness_method = self.fitness
# Clear the best individuals historical cache
self.best_individuals.clear()
def step(self):
offspring = []
while len(offspring) < self.offspring_size:
# Selection
parents = self.selection(self.population, self.selection_size)
# Recombination
if take_chances(self.p_recombination):
progeny = self.recombination(*parents)
else:
progeny = [i.clone() for i in parents]
# Mutation
individuals_who_fit = min(
len(progeny),
self.offspring_size - len(offspring)
)
progeny = [
self.mutation(individual, self.p_mutation)
for individual in random.sample(progeny, individuals_who_fit)
]
# Add progeny to the offspring
offspring.extend(progeny)
# Once offspring is generated, a replace step is performed
self.replacement(self.population, offspring)
# We store the best individual for further information
if self.generation < len(self.best_individuals):
self.best_individuals[self.generation] = self.population.best()
else:
self.best_individuals.append(self.population.best())
def best(self, generation=None):
if self.best_individuals:
generation = generation or -1
if generation > len(self.best_individuals) - 1:
raise PyneticsError()
else:
return self.best_individuals[generation]
else:
return None