def test_incorrect_dimensions_passed_in(self):
def fitness_func(bitstring):
return 0.0
left = Genome(fitness_func, (0, 0, 0, 0, 0))
right = Genome(fitness_func, (1, 1, 1, 1, 1))
dimensions = (4,)
comparer = IncrementalComparer()
rng = NumpyOnesRandom()
genome_builder = UniformCrossoverGenomeFactory(dimensions, rng, 0.0, comparer, fitness_func)
with self.assertRaises(DimensionsMismatchException):
genome_builder.build([left, right])
def test_right_inheritance(self):
def fitness_func(bitstring):
return 0.0
left = Genome(fitness_func, 2 * np.ones((5, 5)))
right = Genome(fitness_func, 3 * np.ones((5, 5)))
dimensions = (5, 5)
comparer = IncrementalComparer()
rng = NumpyPassThroughRandom(0.8)
genome_builder = UniformCrossoverGenomeFactory(dimensions, rng, 0.0, comparer, fitness_func)
new_genome = genome_builder.build([left, right])
self.assertTrue(np.array_equal(right.bitstring, new_genome.bitstring))
def test_mutation(self):
def fitness_func(bitstring):
return 0.0
left = Genome(fitness_func, (0, 0, 0, 0, 0))
right = Genome(fitness_func, (0, 0, 0, 0, 0))
dimensions = (5,)
comparer = IncrementalComparer()
rng = NumpyOnesRandom()
genome_builder = UniformCrossoverGenomeFactory(dimensions, rng, 1.0, comparer, fitness_func)
new_genome = genome_builder.build([left, right])
self.assertEqual([1, 1, 1, 1, 1], list(new_genome.bitstring))
def build(self, prior_genomes: List[Genome]) -> Genome:
""" Builds a new Genome from a list of priors
:param prior_genomes: list of prior Genomes
:return: a new Genome object
"""
def bit_flip(bit_value):
if bit_value == 0:
return 1
else:
return 0
def recursive_bit_flip(bitstring, dimensions):
random_index = self._rng.random_int(0, dimensions[0])
if len(dimensions) == 1:
bitstring[random_index] = bit_flip(bitstring[random_index])
else:
recursive_bit_flip(bitstring[random_index], dimensions[1:])
selected_genome = prior_genomes[self._rng.random_int(
0, len(prior_genomes))]
candidate_genome_bitstring = deepcopy(selected_genome.bitstring)
for _ in range(self._n):
recursive_bit_flip(candidate_genome_bitstring, self._dimensions)
return Genome(self._fitness_func, candidate_genome_bitstring,
*self._args, **self._kwargs)
def test_convergence(self):
def fitness_func_zero(bitstring):
return 0
initial_population = [Genome(fitness_func_zero, (0, ))]
for genome in initial_population:
genome.run()
convergence_criterion = PassThroughConvergenceCriterion(True)
genome_builder = PassThroughGenomeFactory(
Genome(fitness_func_zero, (1, )))
optimizer = BasicGeneticOptimizer(initial_population, genome_builder,
convergence_criterion)
self.assertFalse(optimizer.converged)
optimizer.next()
self.assertTrue(optimizer.converged)
def test_convergence(self):
def fitness_func_zero(bitstring):
return 0
initial_candidate = Genome(fitness_func_zero, (0, ))
initial_candidate.run()
annealing_schedule = PassThroughAnnealingSchedule(0.9, True)
genome_builder = PassThroughGenomeFactory(
Genome(fitness_func_zero, (1, )))
rng = NumpyCustomRandom(0.0, 1)
optimizer = BasicAnnealingOptimizer(initial_candidate, genome_builder,
annealing_schedule, rng)
self.assertFalse(optimizer.converged)
self.assertAlmostEqual(0.9, optimizer.temperature)
optimizer.next()
self.assertTrue(optimizer.converged)
self.assertAlmostEqual(0.9, optimizer.temperature)
def test_next_generation_improvement(self):
def fitness_func_zero(bitstring):
return 0
def fitness_func_one(bitstring):
return 1
initial_population = [Genome(fitness_func_zero, (0, ))]
for genome in initial_population:
genome.run()
convergence_criterion = PassThroughConvergenceCriterion(True)
genome_builder = PassThroughGenomeFactory(
Genome(fitness_func_one, (1, )))
optimizer = BasicGeneticOptimizer(initial_population, genome_builder,
convergence_criterion)
optimizer.next()
self.assertEqual(1, optimizer.best_genome.fitness)
def test_inheritance(self):
def fitness_func(bitstring):
return 0.0
left = Genome(fitness_func, 2 * np.ones((2, 6)))
right = Genome(fitness_func, 3 * np.ones((2, 6)))
dimensions = (2, 6)
comparer = IncrementalComparer()
rng = NumpyPassThroughRandom(3)
genome_builder = SinglePointCrossoverGenomeFactory(
dimensions, rng, 0.0, comparer, fitness_func)
expected = np.ones((2, 6))
expected[:, 3:] = 3
expected[:, :3] = 2
new_genome = genome_builder.build([left, right])
self.assertTrue(np.array_equal(expected, new_genome.bitstring))
def test_worse_solution_acceptance(self):
def fitness_func_zero(bitstring):
return 0
def fitness_func_one(bitstring):
return 1
initial_candidate = Genome(fitness_func_one, (0, ))
initial_candidate.run()
annealing_schedule = PassThroughAnnealingSchedule(0.9, True)
genome_builder = PassThroughGenomeFactory(
Genome(fitness_func_zero, (1, )))
rng = NumpyCustomRandom(0.0, 1)
optimizer = BasicAnnealingOptimizer(initial_candidate, genome_builder,
annealing_schedule, rng)
optimizer.next()
self.assertEqual(0, optimizer.best_genome.fitness)
self.assertAlmostEqual(0.9, optimizer.temperature)
def build(self, prior_genomes: List[Genome]) -> Genome:
""" Builds a new Genome from a list of priors
:param prior_genomes: list of prior Genomes
:return: a new Genome object
"""
def recursive_gene_selection(sub_dimensions, sub_new_bitstring,
sub_left_bitstring, sub_right_bitstring):
if not (sub_dimensions[0] == len(sub_right_bitstring)
and sub_dimensions[0] == len(sub_left_bitstring)):
raise DimensionsMismatchException
if len(sub_dimensions) == 1:
sub_new_bitstring[:
crossover_index] = sub_left_bitstring[:
crossover_index]
sub_new_bitstring[crossover_index:] = sub_right_bitstring[
crossover_index:]
for index in range(sub_dimensions[0]):
if self._rng.random() < self._mutation_rate:
sub_new_bitstring[index] = self._rng.random_int(0, 2)
else:
for index in range(sub_dimensions[0]):
recursive_gene_selection(sub_dimensions[1:],
sub_new_bitstring[index],
sub_left_bitstring[index],
sub_right_bitstring[index])
new_bitstring = self._rng.random_int_array(0, 2, self._dimensions)
left_prior = self._comparer.compare(prior_genomes).bitstring
right_prior = self._comparer.compare(prior_genomes).bitstring
crossover_index = self._rng.random_int(1, self._dimensions[-1])
recursive_gene_selection(self._dimensions, new_bitstring, left_prior,
right_prior)
return Genome(self._fitness_func, new_bitstring, *self._args,
**self._kwargs)