def test_orchestration_run_one_step_missing_data(binh_korn_points):
"""Test that the model also works with missing observations"""
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70])
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1)
palinstance.cross_val_points = 0
# make some of the observations missing
y_binh_korn[:10, 1] = np.nan
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step()
assert idx[0] not in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
def test_crossvalidate(binh_korn_points):
"""Test the crossvalidation routine"""
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70])
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1)
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
original_sample_mask = palinstance.sampled
cross_val_error = palinstance._crossvalidate() # pylint:disable=protected-access
assert (palinstance.sampled_indices == sample_idx).all()
assert (palinstance.sampled == original_sample_mask).all()
assert isinstance(cross_val_error, float)
assert np.abs(cross_val_error) > 0
def test_augment_design_space_bk(binh_korn_points, binh_korn_points_finer):
"""Test the augment function by using a finer sampling of the Binh-Korn function
for augmentation"""
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
(
X_binh_korn_finer, # pylint:disable=invalid-name
_,
) = binh_korn_points_finer
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
new_idx = palinstance.run_one_step()
palinstance.update_train_set(new_idx, y_binh_korn[new_idx])
number_pareto_optimal_points_old = palinstance.number_pareto_optimal_points
palinstance.augment_design_space(X_binh_korn_finer)
assert palinstance.number_discarded_points == 0
assert palinstance.number_pareto_optimal_points > number_pareto_optimal_points_old
def test_pal_sklearn(make_random_dataset):
"""Test that we can create a instanec of the PAL sklearn class"""
X, y = make_random_dataset # pylint:disable=invalid-name
gpr = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
pal_sklearn_instance = PALSklearn(X, [gpr, gpr, gpr], 3)
pal_sklearn_instance.update_train_set(
np.array([1, 2, 3, 4, 5]), y[np.array([1, 2, 3, 4, 5]), :]
)
assert pal_sklearn_instance.models[0].kernel.length_scale == 1
pal_sklearn_instance._train() # pylint:disable=protected-access
assert pal_sklearn_instance.models[0].kernel_.length_scale != 1
def test_gridsearch_object(binh_korn_points):
"""Test the initialization of PALSklearn with a GridsearchCV object"""
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70])
grid_search_0 = GridSearchCV(GaussianProcessRegressor(), {"kernel": [RBF(), Matern()]})
grid_search_1 = GridSearchCV(GaussianProcessRegressor(), {"kernel": [RBF(), Matern()]})
with pytest.raises(ValueError):
palinstance = PALSklearn(X_binh_korn, [grid_search_0, grid_search_1], 2, beta_scale=1)
grid_search_0.fit(X_binh_korn, y_binh_korn[:, 0])
grid_search_1.fit(X_binh_korn, y_binh_korn[:, 1])
palinstance = PALSklearn(X_binh_korn, [grid_search_0, grid_search_1], 2, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step()
assert len(idx) == 1
assert idx[0] not in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
for model in palinstance.models:
assert check_is_fitted(model) is None
def test_orchestration_run_one_step_parallel(binh_korn_points):
"""Test the parallel processing"""
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
gpr_0 = GaussianProcessRegressor(
Matern(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
Matern(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1 / 9, n_jobs=2)
sample_idx = np.array([1, 10, 20, 40, 70, 90])
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
palinstance.cross_val_points = 0
idx = palinstance.run_one_step(batch_size=10)
for index in idx:
assert index not in [1, 10, 20, 40, 70, 90]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
def test_orchestration_run_one_step(make_random_dataset, binh_korn_points):
"""Test if the orchestration works.
In the base class it should raise an error as without
prediction function we cannot do anything
"""
X, y = make_random_dataset # pylint:disable=invalid-name
gpr_0 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
gpr_1 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
gpr_2 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
palinstance = PALSklearn(X, [gpr_0, gpr_1, gpr_2], 3, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y[sample_idx])
idx = palinstance.run_one_step()
assert len(idx) == 1
assert idx[0] not in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
for model in palinstance.models:
assert check_is_fitted(model) is None
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70])
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step()
assert len(idx) == 1
assert idx[0] not in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70]
assert palinstance.number_sampled_points == len(sample_idx)
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
def test_orchestration_run_one_step_batch( # pylint:disable=too-many-statements
binh_korn_points,
):
"""Test the batch sampling"""
X_binh_korn, y_binh_korn = binh_korn_points # pylint:disable=invalid-name
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step(batch_size=10)
assert len(idx) == 10
assert len(np.unique(idx)) == 10
for index in idx:
assert index not in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
# scaling up beta
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1 / 3)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step(batch_size=10)
for index in idx:
assert index not in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 60, 70]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
# smaller initial set
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1 / 3)
palinstance.cross_val_points = 0
sample_idx = np.array([1, 10, 20, 40, 70, 90])
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step(batch_size=10)
for index in idx:
assert index not in [1, 10, 20, 40, 70, 90]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
# smaller initial set and beta scale
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1 / 9)
palinstance.cross_val_points = 0
sample_idx = np.array([1, 10, 20, 40, 70, 90])
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step(batch_size=10)
for index in idx:
assert index not in [1, 10, 20, 40, 70, 90]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
# smaller initial set and beta scale and different kernel
gpr_0 = GaussianProcessRegressor(
Matern(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
gpr_1 = GaussianProcessRegressor(
Matern(), normalize_y=True, n_restarts_optimizer=5, random_state=10
)
palinstance = PALSklearn(X_binh_korn, [gpr_0, gpr_1], 2, beta_scale=1 / 9)
palinstance.cross_val_points = 0
sample_idx = np.array([1, 10, 20, 40, 70, 90])
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step(batch_size=10)
for index in idx:
assert index not in [1, 10, 20, 40, 70, 90]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
# test using the "fixed" epsilon
gpr_0 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=6, random_state=10
)
gpr_1 = GaussianProcessRegressor(
RBF(), normalize_y=True, n_restarts_optimizer=6, random_state=10
)
palinstance = PALSklearn(
X_binh_korn,
[gpr_0, gpr_1],
2,
beta_scale=1 / 9,
ranges=np.ptp(y_binh_korn, axis=0),
)
assert palinstance.uses_fixed_epsilon
palinstance.cross_val_points = 0
sample_idx = np.array([1, 10, 20, 40, 70, 90])
palinstance.update_train_set(sample_idx, y_binh_korn[sample_idx])
idx = palinstance.run_one_step(batch_size=1)
for index in idx:
assert index not in [1, 10, 20, 40, 70, 90]
assert palinstance.number_sampled_points > 0
assert sum(palinstance.unclassified) > 0
assert sum(palinstance.discarded) == 0
for model in palinstance.models:
assert check_is_fitted(model) is None
def test_augment_design_space(make_random_dataset):
"""Test if the reclassification step in the design step
agumentation method works"""
X, y = make_random_dataset # pylint:disable=invalid-name
gpr_0 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
gpr_1 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
gpr_2 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=5)
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
palinstance = PALSklearn(X, [gpr_0, gpr_1, gpr_2], 3, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y[sample_idx])
_ = palinstance.run_one_step()
X_new = X + 1 # pylint:disable=invalid-name
palinstance.augment_design_space(X_new, classify=True, clean_classify=False)
assert palinstance.number_design_points == 200
assert palinstance.number_sampled_points == len(sample_idx)
# Adding new design points should not mess up with the models
for model in palinstance.models:
assert check_is_fitted(model) is None
# Now, test the `clean_classify` flag
gpr_0 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=3)
gpr_1 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=3)
gpr_2 = GaussianProcessRegressor(RBF(), normalize_y=True, n_restarts_optimizer=3)
sample_idx = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
palinstance = PALSklearn(X, [gpr_0, gpr_1, gpr_2], 3, beta_scale=1)
palinstance.cross_val_points = 0
palinstance.update_train_set(sample_idx, y[sample_idx])
_ = palinstance.run_one_step()
X_new = X + np.full((1, 10), 1) # pylint:disable=invalid-name
palinstance.augment_design_space(X_new)
assert palinstance.number_design_points == 200
assert palinstance.number_sampled_points == len(sample_idx)