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_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)