def test_reset_classification(make_random_dataset): """Make sure the reset of the reclassification status works as expected.""" X, _ = make_random_dataset # pylint: disable=invalid-name palinstance = PALBase(X, ["model"], 3) lows = np.zeros((100, 3)) highs = np.zeros((100, 3)) means = np.full((100, 3), 1) palinstance._means = means palinstance.std = np.full((100, 3), 0.1) pareto_optimal = np.array([False] * 98 + [True, True]) sampled = np.array([[False] * 3, [False] * 3, [False] * 3, [False] * 3]) unclassified = np.array([True] * 98 + [False, False]) palinstance.rectangle_lows = lows palinstance.rectangle_ups = highs palinstance.sampled = sampled palinstance.pareto_optimal = pareto_optimal palinstance.unclassified = unclassified palinstance._reset_classification() assert palinstance.number_unclassified_points == len(X) assert palinstance.number_pareto_optimal_points == 0 assert palinstance.number_discarded_points == 0
def test__update_hyperrectangles(make_random_dataset): """Testing if the updating of the hyperrectangles works as expected. As above, the core functionality is tested in for the function with the logic. Here, we're more interested in seeing how it works with the class object """ X, _ = make_random_dataset # pylint:disable=invalid-name palinstance = PALBase(X, ["model"], 4, beta_scale=1) palinstance._means = np.array([[0, 0, 1, 0], [0, 0, 0, 1]]) palinstance.std = np.array([[0, 0, 0, 0], [1, 0, 0, 0]]) with pytest.raises(TypeError): # Beta is not defined palinstance._update_hyperrectangles() palinstance._update_beta() palinstance._update_hyperrectangles() assert palinstance.rectangle_lows is not None assert palinstance.rectangle_ups is not None assert palinstance.rectangle_lows[0][0] == 0 assert palinstance.rectangle_ups[0][0] == 0 assert palinstance.rectangle_lows[0][2] == 1 assert palinstance.rectangle_ups[0][2] == 1 assert palinstance.rectangle_lows[1][0] < -1 assert palinstance.rectangle_ups[1][0] > 1 assert len(palinstance.hyperrectangle_sizes) == len(palinstance._means)
def test__update_coef_var_mask(make_random_dataset): """Test that the coefficient of variation mask works as expected""" X, _ = make_random_dataset # pylint:disable=invalid-name palinstance = PALBase(X[:2], ["model"], 3, beta_scale=1) palinstance._means = np.array([[1, 1, 1, 1], [1, 1, 1, 1]]) palinstance.std = np.array([[0, 0, 0, 0], [3, 1, 1, 1]]) assert (palinstance.coef_var_mask == np.array([True, True])).all() palinstance._update_coef_var_mask() assert (palinstance.coef_var_mask == np.array([True, False])).all() palinstance._means = np.array([[0, 0, 0, 0], [0, 0, 0, 0]]) palinstance.std = np.array([[0, 0, 0, 0], [3, 1, 1, 1]]) assert (palinstance.coef_var_mask == np.array([True, False])).all()
def test__replace_by_measurements(make_random_dataset): """Test that replacing the mean/std by the measured ones works""" X, y = make_random_dataset # pylint:disable=invalid-name palinstance = PALBase(X, ["model"], 3, beta_scale=1) assert palinstance.measurement_uncertainty.sum() == 0 sample_idx = np.array([1, 2, 3, 4]) palinstance.update_train_set(sample_idx, y[sample_idx], y[sample_idx]) palinstance._means = palinstance.measurement_uncertainty palinstance.std = palinstance.measurement_uncertainty palinstance._replace_by_measurements() assert (palinstance.y == palinstance.std).all()
def test_augment_design_space(make_random_dataset): """Testing the basic functionality of the augmentation method Does NOT test the re-classification step, which needs a model""" X, _ = make_random_dataset # pylint: disable=invalid-name X_augmented = np.vstack([X, X]) # pylint: disable=invalid-name palinstance = PALBase(X, ["model"], 3) # Iteration count to low with pytest.raises(ValueError): palinstance.augment_design_space(X_augmented) palinstance.iteration = 2 # Incorrect shape with pytest.raises(AssertionError): palinstance.augment_design_space(X_augmented[:, 2]) with pytest.raises(ValueError): palinstance.augment_design_space(X_augmented[:, :2]) # Mock that we already ran that lows = np.zeros((100, 3)) highs = np.zeros((100, 3)) means = np.full((100, 3), 1) palinstance._means = means palinstance.std = np.full((100, 3), 0.1) pareto_optimal = np.array([False] * 98 + [True, True]) sampled = np.array([[False] * 3, [False] * 3, [False] * 3, [False] * 3]) unclassified = np.array([True] * 98 + [False, False]) palinstance.rectangle_lows = lows palinstance.rectangle_ups = highs palinstance.sampled = sampled palinstance.pareto_optimal = pareto_optimal palinstance.unclassified = unclassified # As we do not have a model, we cannot test the classification palinstance.augment_design_space(X_augmented, clean_classify=False) assert palinstance.number_discarded_points == 0 assert palinstance.number_pareto_optimal_points == 2 assert palinstance.number_unclassified_points == 298 assert palinstance.number_sampled_points == 0 assert palinstance.number_design_points == 300 assert len(palinstance.means) == 300 assert len(palinstance.std) == 300