def test_mutate():
parent = ClassicGRN()
parent.random(5, 5, 5)
config.MUTATION_ADD_RATE = 1.0
child = evolution.mutate(parent)
assert parent.size() == child.size() - 1
assert parent.distance_to(child) < 1 / (child.size() + 2)
config.MUTATION_ADD_RATE = 0.0
config.MUTATION_DEL_RATE = 1.0
child = evolution.mutate(parent)
assert parent.size() == child.size() + 1
assert parent.distance_to(child) < 1 / (child.size() + 2)
config.MUTATION_DEL_RATE = 0.0
child = evolution.mutate(parent)
assert parent.size() == child.size()
assert parent.distance_to(child) < 1 / (child.size() + 2)
def make_offspring(self):
self.offspring = []
for sp in self.species:
if sp.num_offspring == 0:
continue
species_offspring = []
# Create children with crossover
for k in range(int(sp.num_offspring * config.CROSSOVER_RATE)):
parent1 = sp.tournament_select()
parent2 = sp.tournament_select()
# Don't use the same parent 2x
num_tries = 0
while ((parent1 == parent2) and
(num_tries < config.MAX_SELECTION_TRIES)):
parent2 = sp.tournament_select()
num_tries += 1
if parent1 != parent2:
species_offspring += [Individual(
crossover(parent1.grn, parent2.grn))]
# Create children with mutation
for k in range(int(sp.num_offspring * config.MUTATION_RATE)):
parent = sp.tournament_select()
species_offspring += [Individual(mutate(parent.grn))]
# Add elite
if len(species_offspring) == sp.num_offspring:
species_offspring[np.random.randint(
len(species_offspring))] = sp.get_best_individual()
else:
species_offspring += [sp.get_best_individual()]
# Fill with tournament champions if extra space
while len(species_offspring) < sp.num_offspring:
species_offspring += [sp.tournament_select().clone()]
self.offspring += species_offspring
def test_representative_distances():
species = evolution.Species()
assert species.sum_adjusted_fitness is None
problem = Random()
rep = evolution.Individual(ClassicGRN())
rep.grn.random(problem.nin, problem.nout, 1)
species.representative = rep
for i in range(10):
ind = evolution.Individual(evolution.mutate(rep.grn))
species.individuals += [ind]
dists0 = species.get_representative_distances()
assert np.sum(dists0) > 0
rep.grn.random(problem.nin, problem.nout, 1)
dists1 = species.get_representative_distances()
assert np.sum(dists1) > 0
assert np.sum(dists1) > np.sum(dists0)
def speciation(self):
for sp in self.species:
sp.representative = random.choice(sp.individuals).clone()
sp.reset()
for ind in self.offspring:
species_match = None
min_dist = np.inf
for spi in range(len(self.species)):
sp_dist = self.species[spi].representative.grn.distance_to(
ind.grn)
if ((sp_dist < min_dist) and
(sp_dist < self.species[spi].species_threshold)):
species_match = spi
min_dist = sp_dist
if species_match is not None:
self.species[species_match].individuals += [ind]
else:
new_sp = Species()
new_sp.individuals += [ind]
new_sp.representative = ind.clone()
self.species += [new_sp]
self.offspring = []
num_removed = 0
species_to_remove = []
for sp in self.species:
if len(sp.individuals) < config.MIN_SPECIES_SIZE:
num_removed += len(sp.individuals)
species_to_remove += [sp]
del sp.representative
for ind in sp.individuals:
del ind
self.species = list(set(self.species) - set(species_to_remove))
if len(self.species) == 0:
raise Exception(('The entire population was removed. '
'Try again with a smaller MIN_SPECIES_SIZE '
'or a larger INITIALIZATION_DUPLICATION'))
while num_removed > 0:
sp = random.choice(self.species)
ind = sp.tournament_select()
tries = 0
has_added = False
while not has_added:
child_grn = mutate(ind.grn)
if (child_grn.distance_to(sp.representative.grn) <
sp.species_threshold):
sp.individuals += [Individual(child_grn)]
num_removed -= 1
has_added = True
tries += 1
if tries == config.MAX_SELECTION_TRIES:
ind = Individual(child_grn)
new_sp = Species()
new_sp.individuals += [ind]
new_sp.representative = ind.clone()
self.species += [new_sp]
num_removed -= 1
has_added = True