def crossover_intermediate(population, alpha=None):
"""
Generated children feature averages of the equivalent parent gene elements.
"""
assert isinstance(population, Population)
assert isinstance(population.individuals[0].gene[0], (bool, int, float))
assert len(population) == 2, "this works with two parents, only."
# TODO: How about a universal formula (for p > 2)?
gene = []
for gene_index in range(population.gene_size):
features_parents = [parent.gene[gene_index]
for parent in population.individuals]
if alpha is None:
# calculate average feature
feature = statistics.mean(features_parents)
# cast back to original type
feature = type(population.individuals[0].gene[0])(feature)
else:
# reweight according to alpha
feature = sum([alpha * features_parents[0],
(1 - alpha) * features_parents[1]])
print(feature)
gene.append(feature)
return Population.from_individuals(individuals=[Individual(gene=gene)])
def crossover_n_point(population, n=1):
assert isinstance(population, Population)
# select crossover point
valid_positions = range(1, population.gene_size-1)
points = [random.choice(valid_positions) for _ in range(n)]
# generate children from parents
children = []
# iterate over children
for i in range(len(population)):
gene_child = []
# iterate over crossover points
for i in range(-1, n):
if i == -1:
# part left of the first crossover point
gene_child += population[0].gene[0:points[0]]
continue
key = i % population.gene_size - 1
parent = population[key]
try:
gene_child += parent.gene[points[i]:points[i+1]]
except IndexError:
# this is the last point
gene_child += parent.gene[points[i]:]
child_individual = Individual(gene=gene_child)
children.append(child_individual)
return Population.from_individuals(children)
def ea(SIGMA=SIGMA):
# initialize
world = World(fitness=fitness)
world.population = IntListPopulation(length=MU, gene_size=GENE_SIZE)
# iterate over generations
for generation_count in range(MAX_GENERATIONS):
# log this generation's population to stdout
print("%s: %s" % (generation_count, world.population))
# if we reached a sufficiently good solution, stop
if min(world.scores) < MIN_FITNESS:
break
# select rho fittest individuals as parents
parents = select(world=world, rho=RHO)
# generate children by crossing over the parents
children_original = []
while len(children_original) < LAMBDA:
children_original += crossover(population=parents)
# mutate children
children = [mutate(individual=individual, mu=0, sigma=SIGMA)
for individual in children_original]
# self-adapt mutation strenght using Rechenberg's success rule
SIGMA = rechenberg(children_original=children_original,
children_mutated=children, fitness=world.fitness,
a=A_RECHENBERG, sigma=SIGMA)
if MODE == '+':
# extend population with children
world.population += Population.from_individuals(
individuals=children)
elif MODE == ',':
# replace population with children
world.population = Population.from_individuals(
individuals=children)
# select fittest mu individuals for next generation
world.population = select(rho=MU, world=world)
# log best individual
print("Best individual is %s with a score of %s." %
(select(world=world, rho=1)[0], min(world.scores)))
def crossover_mode_bool(population):
gene = []
for i in range(len(population[0].gene)):
base = sum([1 for parent in population if parent.gene[i-1]]) > \
len(population) / 2
gene.append(base)
individual = Individual(gene=gene)
return Population.from_individuals(individuals=[individual])
def crossover_dominant(population):
"""
Collects child gene from equivalent gene elements of a random parent.
"""
assert isinstance(population, Population)
gene = []
for gene_index in range(population.gene_size):
parent_index = random.choice(range(len(population)))
gene.append(population[parent_index].gene[gene_index])
return Population.from_individuals(individuals=[Individual(gene=gene)])
