def test_crossover_allows_to_change_last_element(self, cfg):
# given
cl1 = Classifier(condition='0##10###', cfg=cfg)
cl2 = Classifier(condition='#10##011', cfg=cfg)
# when
two_point_crossover(cl1, cl2, samplefunc=SampleMock([5, 8]))
# then
assert Condition('0##10011') == cl1.condition
assert Condition('#10#####') == cl2.condition
def test_crossover(self, cfg):
# given
cl1 = Classifier(condition='0##10###', cfg=cfg)
cl2 = Classifier(condition='#10##0##', cfg=cfg)
# when
two_point_crossover(cl1, cl2, samplefunc=SampleMock([1, 4]))
# then
assert Condition('010#0###') == cl1.condition
assert Condition('###1#0##') == cl2.condition
def apply_ga(self,
time: int,
population: ClassifiersList,
match_set: ClassifiersList,
situation: Perception,
randomfunc=random,
samplefunc=sample) -> None:
if self.should_apply_ga(time):
self.set_ga_timestamp(time)
parent1, parent2 = roulette_wheel_parents_selection(
self, randomfunc=randomfunc)
child1 = Classifier.copy_from(parent1, time)
child2 = Classifier.copy_from(parent2, time)
mutate(child1, child1.cfg.mu, randomfunc=randomfunc)
mutate(child2, child2.cfg.mu, randomfunc=randomfunc)
if randomfunc() < self.cfg.chi:
if child1.effect == child2.effect:
two_point_crossover(child1, child2, samplefunc=samplefunc)
# Update quality and reward
# TODO: check if needed
child2.q = float(sum([child1.q, child2.q]) / 2)
child2.r = float(sum([child1.r, child2.r]) / 2)
child1.q /= 2
child2.q /= 2
children = [
child for child in [child1, child2]
if child.condition.specificity > 0
]
# if two classifiers are identical, leave only one
unique_children = set(children)
self.delete_ga_classifiers(population,
match_set,
len(unique_children),
randomfunc=randomfunc)
# check for subsumers / similar classifiers
for child in unique_children:
self.add_ga_classifier(child, match_set, population)