def test_parameter_passing_e2e(self):
class MyMember(ga.MemberBase):
def _construct_from_params(self, construction_parameters=None):
self.height = construction_parameters['height']
self.width = construction_parameters['width']
def mutate(self):
i = random.randrange(2)
j = random.randrange(2)
self.height += pow(-1, i)
self.width += pow(-1, j)
def crossover(self, pairing):
new_params = {
'height': np.mean([self.height, pairing.height]),
'width': np.mean([self.width, pairing.width])
}
return MyMember(construction_parameters=new_params)
def fit_through_door(member: MyMember):
door_height = 10
door_width = 3
height_diff = abs(door_height - member.height)
width_diff = abs(door_width - member.width)
return height_diff + width_diff
def param_generator():
max_h = 10
max_w = 20
yield {
'height': random.randrange(max_h),
'width': random.randrange(max_w)
}
random.seed(1)
pop = ga.Population(10,
MyMember,
fit_through_door,
member_parameters_generator=param_generator)
self.assertEqual(len(pop.population), 10)
self.assertEqual(pop.member_type, MyMember)
self.assertEqual(pop.fitness_function, fit_through_door)
self.assertDictEqual(pop.population[0].construction_parameters, {
'height': 2,
'width': 18
})
# run 5 generations before checking we have some parameters changed
pop.run(generations=500, maximise_fitness_func=False)
self.assertEqual({
'height': 9.6953125,
'width': 4.1953125
}, pop.population[0].construction_parameters)
def test_creation_basic(self):
class MyMember(ga.MemberBase):
def _construct_from_params(self, construction_parameters=None):
self.vars = [random.randrange(10) for _ in range(10)]
# Create mutate and score are not needed for creation
def mutate(self):
pass
def crossover(self, pairing):
pass
def score_self(self):
pass
pop = ga.Population(10, MyMember, MyMember.score_self)
self.assertEqual(len(pop.population), 10)
self.assertEqual(pop.member_type, MyMember)
self.assertEqual(pop.fitness_function, MyMember.score_self)
def test_e2e_basic(self):
class MyMember(ga.MemberBase):
def _construct_from_params(self, construction_parameters=None):
self.vars = [random.randrange(10) for _ in range(10)]
def mutate(self):
j = random.randrange(len(self.vars))
self.vars[j] += 1
def crossover(self, pairing):
new_params = self.vars[:len(self.vars) //
2] + pairing.vars[len(pairing.vars) //
2:]
return MyMember(new_params)
def score_self(self):
return sum(self.vars)
def m_fit_func(member: MyMember):
return member.score_self()
pop = ga.Population(size=10,
member_type=MyMember,
member_parameters_generator=None,
fitness_function=m_fit_func,
population_seed=0)
start_pop = pop.population
start_scores = list(starmap(m_fit_func, [[m] for m in start_pop]))
# Ensure good creation
self.assertEqual(len(pop.population), 10)
self.assertEqual(pop.member_type, MyMember)
self.assertEqual(pop.fitness_function, m_fit_func)
pop.run(10, print_logging=False, maximise_fitness_func=True)
end_scores = list(starmap(m_fit_func, [[m] for m in pop.population]))
# Ensure changes have been made
self.assertNotEqual(pop.population, start_pop)
self.assertGreaterEqual(max(end_scores), max(start_scores))
class MyMember(ga.MemberBase):
def _construct_from_params(self, construction_parameters=None):
self.vars = [random.randrange(10) for _ in range(5)]
def mutate(self):
# i = random.randrange(len(self.vars))
j = random.randrange(len(self.vars))
# tmp = self.vars[i]
# self.vars[i] = self.vars[j]
self.vars[j] += 1
def crossover(self, pairing):
new_params = self.vars[:len(self.vars) //
2] + pairing.vars[len(pairing.vars) // 2:]
return MyMember(new_params)
def score_self(self):
return sum(self.vars)
def m_fit_func(member: MyMember):
return member.score_self()
pop = ga.Population(size=10,
member_type=MyMember,
member_parameters_generator=None,
fitness_function=m_fit_func,
population_seed=0)
pop.run(100, print_logging=True, csv_path="example1.csv")
if height_diff > 0 and width_diff > 0:
return (1 / height_diff) + (1 / width_diff)
elif height_diff == 0:
return 1 / width_diff
elif width_diff == 0:
return 1 / height_diff
else:
return np.inf
def param_generator():
max_h = 10
max_w = 20
yield {'height': random.randrange(max_h), 'width': random.randrange(max_w)}
random.seed(1)
pop = ga.Population(100,
Door,
fit_through_door,
member_parameters_generator=param_generator)
# run 500 generations before checking we have some parameters changed
pop.run(
generations=500,
print_logging=True,
)
print("Best door:", pop.get_top())
# This will spit out a very close approximation of the true door size we selected