def setUp(self):
rng = np.random.RandomState(42)
Cluster.global_rng = rng
Genotype.global_rng = rng
setattr(Cluster, "num_dims", 2)
clust1 = Cluster(50)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[1, 0], [0, 1]])
clust2 = Cluster(30)
clust2.mean = np.array([5, 5])
clust2.cov = np.array([[5, 0], [0, 10]])
self.indiv1 = Genotype([clust1, clust2])
self.indiv1.create_views()
self.indiv1.resample_values()
clust3 = Cluster(50)
clust3.mean = np.array([2, 2])
clust3.cov = np.array([[2, 0], [0, 2]])
clust4 = Cluster(30)
clust4.mean = np.array([10, 10])
clust4.cov = np.array([[4, 0], [0, 2]])
self.indiv2 = Genotype([clust3, clust4])
self.indiv2.create_views()
self.indiv2.resample_values()
def setUp(self):
# Whatever setup is needed
rng = np.random.RandomState(42)
Genotype.global_rng = rng
Cluster.global_rng = rng
sizes = [190, 20, 30, 110]
self.indiv = Genotype([Cluster(size) for size in sizes])
self.indiv.create_views()
self.indiv.resample_values()
hawks.objectives.Silhouette.setup_indiv(self.indiv)
def setup_ga(ga_params, constraint_params, objective_params, dataset_obj):
# Validate the constraints parameters
Genotype.validate_constraints(constraint_params)
# Setup the objective parameters
objective_dict, ga_params = _setup_objectives(objective_params, ga_params)
# Create the DEAP toolbox and generate the initial population
toolbox, initial_pop = ga.main_setup(
objective_dict, dataset_obj, ga_params, constraint_params
)
return objective_dict, ga_params, toolbox, initial_pop
def setUp(self):
# Whatever setup is needed
rng = np.random.RandomState(42)
Genotype.global_rng = rng
Cluster.global_rng = rng
sizes = [190, 20, 30, 110]
setattr(Cluster, "num_dims", 2)
setattr(Cluster, "initial_mean_upper", 1.0)
setattr(Cluster, "initial_cov_upper", 0.5)
self.indiv = Genotype([Cluster(size) for size in sizes])
self.indiv.create_views()
self.indiv.resample_values()
hawks.objectives.Silhouette.setup_indiv(self.indiv)
def setup_ga(self, dataset_obj, config):
# Validate the constraints parameters
Genotype.validate_constraints(config["constraints"])
# Setup the objective parameters
objective_dict = self._setup_objectives(config)
# Create the DEAP toolbox and generate the initial population
toolbox, initial_pop = ga.main_setup(
objective_dict=objective_dict,
dataset_obj=dataset_obj,
ga_params=config["ga"],
constraint_params=config["constraints"]
)
# Set the DEAP toolbox as an attr
self.deap_toolbox = toolbox
# Return the objective dict and the initial population
return objective_dict, initial_pop
def test_silhouette_singleton_cluster(self):
rng = np.random.RandomState(42)
Genotype.global_rng = rng
Cluster.global_rng = rng
sizes = [1, 20, 30, 110]
self.indiv = Genotype([Cluster(size) for size in sizes])
self.indiv.create_views()
self.indiv.resample_values()
hawks.objectives.Silhouette.setup_indiv(self.indiv)
hawks.objectives.Silhouette.eval_objective(self.indiv)
close_to_sk = np.isclose(
silhouette_score(self.indiv.all_values,
self.indiv.labels,
metric="sqeuclidean"), self.indiv.silhouette)
self.assertTrue(close_to_sk)
def test_views_swap_all_genes(self):
orig_all_vals = np.copy(self.indiv2.all_values)
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
mixing_ratio=1.0)
self.indiv1.recreate_views()
self.indiv1.resample_values()
close = np.allclose(self.indiv1.all_values, orig_all_vals)
self.assertTrue(close)
def test_views_swap_all_clusters(self):
orig_all_vals = np.copy(self.indiv2.all_values)
self.indiv1, self.indiv2 = Genotype.xover_cluster(self.indiv1,
self.indiv2,
cxpb=1)
self.indiv1.recreate_views()
self.indiv1.resample_values()
close = np.allclose(self.indiv1.all_values, orig_all_vals)
self.assertTrue(close)
def test_eigenratio_eccentric(self):
clust1 = Cluster(50)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[10, 0], [0, 1]])
clust2 = Cluster(80)
clust2.mean = np.array([0, 0])
clust2.cov = np.array([[9.9, 0], [0, 1]])
indiv = Genotype([clust1, clust2])
eigen_ratio = hawks.constraints.eigenval_ratio(indiv)
self.assertEqual(eigen_ratio, 10)
def test_uniform_crossover_none(self):
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
mixing_ratio=0.0)
indiv1_unchanged = all([
np.allclose(self.indiv1[0].mean, np.array([0, 0])),
np.allclose(self.indiv1[0].cov, np.array([[1, 0], [0, 1]])),
np.allclose(self.indiv1[1].mean, np.array([5, 5])),
np.allclose(self.indiv1[1].cov, np.array([[5, 0], [0, 10]]))
])
self.assertTrue(indiv1_unchanged)
def test_uniform_crossover_all(self):
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
cxpb=1)
indiv1_allchanged = not any([
np.allclose(self.indiv1[0].mean, np.array([0, 0])),
np.allclose(self.indiv1[0].cov, np.array([[1, 0], [0, 1]])),
np.allclose(self.indiv1[1].mean, np.array([5, 5])),
np.allclose(self.indiv1[1].cov, np.array([[5, 0], [0, 10]]))
])
self.assertTrue(indiv1_allchanged)
def test_uniform_crossover_genes(self):
# Get the sequence of the numbers to be generated
rng = np.random.RandomState(42)
swaps = [
True if rng.rand() < 0.5 else False
for _ in range(len(self.indiv1) * 2)
]
indiv1_means = [i.mean for i in self.indiv1]
indiv2_means = [i.mean for i in self.indiv2]
indiv1_covs = [i.cov for i in self.indiv1]
indiv2_covs = [i.cov for i in self.indiv2]
Genotype.global_rng = np.random.RandomState(42)
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
mixing_ratio=0.5)
# Test means
for i, (clust1, clust2,
swap) in enumerate(zip(self.indiv1, self.indiv2, swaps[0::2])):
with self.subTest(i=i):
if swap:
self.assertTrue(
np.array_equal(clust1.mean, indiv2_means[i]))
self.assertTrue(
np.array_equal(clust2.mean, indiv1_means[i]))
else:
self.assertTrue(
np.array_equal(clust1.mean, indiv1_means[i]))
self.assertTrue(
np.array_equal(clust2.mean, indiv2_means[i]))
# Test covs
for i, (clust1, clust2,
swap) in enumerate(zip(self.indiv1, self.indiv2, swaps[1::3])):
with self.subTest(i=i):
if swap:
self.assertTrue(np.array_equal(clust1.cov, indiv2_covs[i]))
self.assertTrue(np.array_equal(clust2.cov, indiv1_covs[i]))
else:
self.assertTrue(np.array_equal(clust1.cov, indiv1_covs[i]))
self.assertTrue(np.array_equal(clust2.cov, indiv2_covs[i]))
def test_uniform_crossover_clusters(self):
rng = np.random.RandomState(42)
swaps = [True if rng.rand() < 0.5 else False for _ in self.indiv1]
indiv1_ids = [id(i) for i in self.indiv1]
indiv2_ids = [id(i) for i in self.indiv2]
Genotype.global_rng = np.random.RandomState(42)
self.indiv1, self.indiv2 = Genotype.xover_cluster(
self.indiv1, self.indiv2)
# Test whether cluster objects have been swapped or not
for i, (clust1, clust2,
swap) in enumerate(zip(self.indiv1, self.indiv2, swaps)):
with self.subTest(i=i):
if swap:
self.assertEqual(indiv1_ids[i], id(clust2))
self.assertEqual(indiv2_ids[i], id(clust1))
else:
self.assertEqual(indiv1_ids[i], id(clust1))
self.assertEqual(indiv2_ids[i], id(clust2))
def test_overlap_same(self):
clust1 = Cluster(4000)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[1, 0], [0, 1]])
clust2 = Cluster(4000)
clust2.mean = np.array([0, 0])
clust2.cov = np.array([[1, 0], [0, 1]])
indiv = Genotype([clust1, clust2])
indiv.create_views()
indiv.resample_values()
overlap = hawks.constraints.overlap(indiv)
self.assertAlmostEqual(overlap, 0.5, places=1)
def test_overlap_separated(self):
clust1 = Cluster(50)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[1, 0], [0, 1]])
clust2 = Cluster(80)
clust2.mean = np.array([10, 10])
clust2.cov = np.array([[1, 0], [0, 1]])
indiv = Genotype([clust1, clust2])
indiv.create_views()
indiv.resample_values()
overlap = hawks.constraints.overlap(indiv)
self.assertEqual(overlap, 0)
class GenotypeTests(unittest.TestCase):
# Test some methods of the Genotype - mainly the array view manipulation
def setUp(self):
rng = np.random.RandomState(42)
Cluster.global_rng = rng
Genotype.global_rng = rng
setattr(Cluster, "num_dims", 2)
clust1 = Cluster(50)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[1, 0], [0, 1]])
clust2 = Cluster(30)
clust2.mean = np.array([5, 5])
clust2.cov = np.array([[5, 0], [0, 10]])
self.indiv1 = Genotype([clust1, clust2])
self.indiv1.create_views()
self.indiv1.resample_values()
clust3 = Cluster(50)
clust3.mean = np.array([2, 2])
clust3.cov = np.array([[2, 0], [0, 2]])
clust4 = Cluster(30)
clust4.mean = np.array([10, 10])
clust4.cov = np.array([[4, 0], [0, 2]])
self.indiv2 = Genotype([clust3, clust4])
self.indiv2.create_views()
self.indiv2.resample_values()
def tearDown(self):
Cluster.global_rng = None
Genotype.global_rng = None
delattr(Cluster, "num_dims")
def test_views_swap_all_genes(self):
orig_all_vals = np.copy(self.indiv2.all_values)
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
cxpb=1)
self.indiv1.recreate_views()
self.indiv1.resample_values()
close = np.allclose(self.indiv1.all_values, orig_all_vals)
self.assertTrue(close)
def test_views_swap_all_clusters(self):
orig_all_vals = np.copy(self.indiv2.all_values)
self.indiv1, self.indiv2 = Genotype.xover_cluster(self.indiv1,
self.indiv2,
cxpb=1)
self.indiv1.recreate_views()
self.indiv1.resample_values()
close = np.allclose(self.indiv1.all_values, orig_all_vals)
self.assertTrue(close)
class ObjectiveTests(unittest.TestCase):
# Test that the objectives return as they should
def setUp(self):
# Whatever setup is needed
rng = np.random.RandomState(42)
Genotype.global_rng = rng
Cluster.global_rng = rng
sizes = [190, 20, 30, 110]
self.indiv = Genotype([Cluster(size) for size in sizes])
self.indiv.create_views()
self.indiv.resample_values()
hawks.objectives.Silhouette.setup_indiv(self.indiv)
def tearDown(self):
self.indiv = None
rng = None
Genotype.global_rng = None
Cluster.global_rng = None
def test_silhouette_complete_indiv(self):
hawks.objectives.Silhouette.eval_objective(self.indiv)
close_to_sk = np.isclose(
silhouette_score(self.indiv.all_values,
self.indiv.labels,
metric="sqeuclidean"), self.indiv.silhouette)
self.assertTrue(close_to_sk)
def test_silhouette_changed_indiv(self):
hawks.objectives.Silhouette.eval_objective(self.indiv)
for cluster in self.indiv:
cluster.changed = False
self.indiv[0].gen_initial_mean()
self.indiv[0].changed = True
aux = self.indiv.silhouette
self.indiv.resample_values()
hawks.objectives.Silhouette.eval_objective(self.indiv)
self.assertNotAlmostEqual(aux, self.indiv.silhouette)
close_to_sk = np.isclose(
silhouette_score(self.indiv.all_values,
self.indiv.labels,
metric="sqeuclidean"), self.indiv.silhouette)
self.assertTrue(close_to_sk)
def test_silhouette_singleton_cluster(self):
rng = np.random.RandomState(42)
Genotype.global_rng = rng
Cluster.global_rng = rng
sizes = [1, 20, 30, 110]
self.indiv = Genotype([Cluster(size) for size in sizes])
self.indiv.create_views()
self.indiv.resample_values()
hawks.objectives.Silhouette.setup_indiv(self.indiv)
hawks.objectives.Silhouette.eval_objective(self.indiv)
close_to_sk = np.isclose(
silhouette_score(self.indiv.all_values,
self.indiv.labels,
metric="sqeuclidean"), self.indiv.silhouette)
self.assertTrue(close_to_sk)
class OperatorTests(unittest.TestCase):
# Test mutation and crossover operators
def setUp(self):
rng = np.random.RandomState(42)
Cluster.global_rng = rng
Genotype.global_rng = rng
setattr(Cluster, "num_dims", 2)
setattr(Cluster, "initial_mean_upper", 1.0)
setattr(Cluster, "initial_cov_upper", 0.5)
clust1 = Cluster(70)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[1, 0], [0, 1]])
clust2 = Cluster(90)
clust2.mean = np.array([5, 5])
clust2.cov = np.array([[5, 0], [0, 10]])
self.indiv1 = Genotype([clust1, clust2])
self.indiv1.create_views()
self.indiv1.resample_values()
clust3 = Cluster(70)
clust3.mean = np.array([2, 2])
clust3.cov = np.array([[2, 0], [0, 2]])
clust4 = Cluster(90)
clust4.mean = np.array([10, 10])
clust4.cov = np.array([[4, 0], [0, 2]])
self.indiv2 = Genotype([clust3, clust4])
self.indiv2.create_views()
self.indiv2.resample_values()
def tearDown(self):
Cluster.global_rng = None
Genotype.global_rng = None
delattr(Cluster, "num_dims")
def test_uniform_crossover_genes(self):
# Get the sequence of the numbers to be generated
rng = np.random.RandomState(42)
swaps = [
True if rng.rand() < 0.5 else False
for _ in range(len(self.indiv1) * 2)
]
indiv1_means = [i.mean for i in self.indiv1]
indiv2_means = [i.mean for i in self.indiv2]
indiv1_covs = [i.cov for i in self.indiv1]
indiv2_covs = [i.cov for i in self.indiv2]
Genotype.global_rng = np.random.RandomState(42)
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
mixing_ratio=0.5)
# Test means
for i, (clust1, clust2,
swap) in enumerate(zip(self.indiv1, self.indiv2, swaps[0::2])):
with self.subTest(i=i):
if swap:
self.assertTrue(
np.array_equal(clust1.mean, indiv2_means[i]))
self.assertTrue(
np.array_equal(clust2.mean, indiv1_means[i]))
else:
self.assertTrue(
np.array_equal(clust1.mean, indiv1_means[i]))
self.assertTrue(
np.array_equal(clust2.mean, indiv2_means[i]))
# Test covs
for i, (clust1, clust2,
swap) in enumerate(zip(self.indiv1, self.indiv2, swaps[1::3])):
with self.subTest(i=i):
if swap:
self.assertTrue(np.array_equal(clust1.cov, indiv2_covs[i]))
self.assertTrue(np.array_equal(clust2.cov, indiv1_covs[i]))
else:
self.assertTrue(np.array_equal(clust1.cov, indiv1_covs[i]))
self.assertTrue(np.array_equal(clust2.cov, indiv2_covs[i]))
def test_uniform_crossover_clusters(self):
rng = np.random.RandomState(42)
swaps = [True if rng.rand() < 0.5 else False for _ in self.indiv1]
indiv1_ids = [id(i) for i in self.indiv1]
indiv2_ids = [id(i) for i in self.indiv2]
Genotype.global_rng = np.random.RandomState(42)
self.indiv1, self.indiv2 = Genotype.xover_cluster(
self.indiv1, self.indiv2)
# Test whether cluster objects have been swapped or not
for i, (clust1, clust2,
swap) in enumerate(zip(self.indiv1, self.indiv2, swaps)):
with self.subTest(i=i):
if swap:
self.assertEqual(indiv1_ids[i], id(clust2))
self.assertEqual(indiv2_ids[i], id(clust1))
else:
self.assertEqual(indiv1_ids[i], id(clust1))
self.assertEqual(indiv2_ids[i], id(clust2))
def test_uniform_crossover_none(self):
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
mixing_ratio=0.0)
indiv1_unchanged = all([
np.allclose(self.indiv1[0].mean, np.array([0, 0])),
np.allclose(self.indiv1[0].cov, np.array([[1, 0], [0, 1]])),
np.allclose(self.indiv1[1].mean, np.array([5, 5])),
np.allclose(self.indiv1[1].cov, np.array([[5, 0], [0, 10]]))
])
self.assertTrue(indiv1_unchanged)
def test_uniform_crossover_all(self):
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
mixing_ratio=1.0)
indiv1_allchanged = not any([
np.allclose(self.indiv1[0].mean, np.array([0, 0])),
np.allclose(self.indiv1[0].cov, np.array([[1, 0], [0, 1]])),
np.allclose(self.indiv1[1].mean, np.array([5, 5])),
np.allclose(self.indiv1[1].cov, np.array([[5, 0], [0, 10]]))
])
self.assertTrue(indiv1_allchanged)
class OperatorTests(unittest.TestCase):
# Test mutation and crossover operators
def setUp(self):
rng = np.random.RandomState(42)
Cluster.global_rng = rng
Genotype.global_rng = rng
setattr(Cluster, "num_dims", 2)
clust1 = Cluster(70)
clust1.mean = np.array([0, 0])
clust1.cov = np.array([[1, 0], [0, 1]])
clust2 = Cluster(90)
clust2.mean = np.array([5, 5])
clust2.cov = np.array([[5, 0], [0, 10]])
self.indiv1 = Genotype([clust1, clust2])
self.indiv1.create_views()
self.indiv1.resample_values()
clust3 = Cluster(70)
clust3.mean = np.array([2, 2])
clust3.cov = np.array([[2, 0], [0, 2]])
clust4 = Cluster(90)
clust4.mean = np.array([10, 10])
clust4.cov = np.array([[4, 0], [0, 2]])
self.indiv2 = Genotype([clust3, clust4])
self.indiv2.create_views()
self.indiv2.resample_values()
def tearDown(self):
Cluster.global_rng = None
Genotype.global_rng = None
delattr(Cluster, "num_dims")
def test_uniform_crossover_none(self):
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
cxpb=0)
indiv1_unchanged = all([
np.allclose(self.indiv1[0].mean, np.array([0, 0])),
np.allclose(self.indiv1[0].cov, np.array([[1, 0], [0, 1]])),
np.allclose(self.indiv1[1].mean, np.array([5, 5])),
np.allclose(self.indiv1[1].cov, np.array([[5, 0], [0, 10]]))
])
self.assertTrue(indiv1_unchanged)
def test_uniform_crossover_all(self):
self.indiv1, self.indiv2 = Genotype.xover_genes(self.indiv1,
self.indiv2,
cxpb=1)
indiv1_allchanged = not any([
np.allclose(self.indiv1[0].mean, np.array([0, 0])),
np.allclose(self.indiv1[0].cov, np.array([[1, 0], [0, 1]])),
np.allclose(self.indiv1[1].mean, np.array([5, 5])),
np.allclose(self.indiv1[1].cov, np.array([[5, 0], [0, 10]]))
])
self.assertTrue(indiv1_allchanged)