def test_preprocess_dsel_scores():
ds_test = DS(create_pool_classifiers())
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
dsel_scores = ds_test._preprocess_dsel_scores()
expected = np.array([[0.5, 0.5], [1.0, 0.0], [0.33, 0.67]])
expected = np.tile(expected, (15, 1, 1))
assert np.array_equal(dsel_scores, expected)
def test_all_classifiers_agree():
# 10 classifiers that return 1
pool_classifiers = [create_base_classifier(return_value=1)] * 10
ds = DS(pool_classifiers)
x = np.ones((1, 10))
assert ds._all_classifier_agree(x)
def test_not_all_classifiers_agree():
# 10 classifiers that return 1, and one that returns 2
pool_classifiers = [create_base_classifier(return_value=1)] * 10
pool_classifiers.append(create_base_classifier(return_value=2))
ds = DS(pool_classifiers)
x = np.ones((1, 10))
assert not ds._all_classifier_agree(x)
def test_instance_hardness_region_all_same():
X = X_dsel_ex1
y = y_dsel_ex1
ds_test = DS(create_pool_classifiers())
ds_test.fit(X, y)
neighbors = [0, 1, 2, 6, 7, 8, 13]
IH = ds_test._hardness_region_competence(neighbors)
assert IH == 0.0
def test_instance_hardness_region(index, expected):
X = X_dsel_ex1
y = y_dsel_ex1
ds_test = DS(create_pool_classifiers())
ds_test.fit(X, y)
neighbors = neighbors_ex1[index, :]
IH = ds_test._hardness_region_competence(neighbors)
assert np.isclose(IH, expected, atol=0.01)
def test_IH_is_used(index, expected):
query = np.atleast_2d([1, 0])
ds_test = DS(create_pool_classifiers(), with_IH=True, IH_rate=0.5)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.processed_dsel = dsel_processed_ex1
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
ds_test.neighbors = neighbors_ex1[index, :]
ds_test.distances = distances_ex1[index, :]
predicted = ds_test.predict(query)
assert predicted == expected
def test_IH_is_used():
expected = [0, 0, 1]
query = np.ones((3, 2))
ds_test = DS(create_pool_classifiers(), with_IH=True, IH_rate=0.5)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.processed_dsel = dsel_processed_ex1
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
ds_test.neighbors = neighbors_ex1
ds_test.distances = distances_ex1
predicted = ds_test.predict(query)
assert np.array_equal(predicted, expected)
def test_label_encoder_base():
from sklearn.linear_model import LogisticRegression
X_dsel_ex1 = np.array([[-1, 1], [-0.75, 0.5], [-1.5, 1.5]])
y_dsel_ex1 = np.array(['cat', 'dog', 'plane'])
x = [[-2, -2], [2, 2]]
y = ['cat', 'dog']
pool_classifiers = [LogisticRegression().fit(x, y) for _ in range(2)]
query = np.atleast_2d([[1, 0], [-1, -1]])
ds_test = DS(pool_classifiers, k=2)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
predictions = ds_test.predict(query)
assert np.equal(predictions, ['cat', 'dog'])
def test_get_dsel_scores():
ds_test = DS(create_pool_classifiers())
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.dsel_scores = dsel_scores_ex1
assert np.array_equal(ds_test._get_scores_dsel(0, 0), np.array([1.0, 0.0]))
assert np.array_equal(ds_test._get_scores_dsel(1, 0), np.array([0.5, 0.5]))
assert np.array_equal(ds_test._get_scores_dsel(2, 0), np.array([0.8, 0.2]))
def test_predict_proba_exists():
query = np.array([[1.0, 1.0]])
X = X_dsel_ex1
y = y_dsel_ex1
clf1 = Perceptron()
clf1.fit(X, y)
with pytest.raises(ValueError):
DS([clf1, clf1])
def test_predict_proba_instance_called(index):
query = np.atleast_2d([1, 1])
ds_test = DS(create_pool_classifiers(), with_IH=True, IH_rate=0.10)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.neighbors = neighbors_ex1[index, :]
ds_test.distances = distances_ex1[index, :]
ds_test.predict_proba_instance = MagicMock(return_value=np.atleast_2d([0.25, 0.75]))
proba = ds_test.predict_proba(query)
assert np.isclose(proba, np.atleast_2d([0.25, 0.75])).all()
def test_frienemy_all_classifiers_crosses(index):
ds_test = DS(create_pool_classifiers())
ds_test.processed_dsel = dsel_processed_all_ones
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
ds_test.neighbors = neighbors_ex1[index, :]
result = ds_test._frienemy_pruning()
assert result.all() == 1.0
def test_predict_proba_all_agree():
query = np.atleast_2d([1, 1])
ds_test = DS(create_pool_classifiers())
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.dsel_scores = dsel_scores_ex1
ds_test._all_classifier_agree = MagicMock(return_value=True)
proba = ds_test.predict_proba(query)
assert np.isclose(proba, np.atleast_2d([0.61, 0.39])).all()
def test_frienemy_safe_region():
ds_test = DS(create_pool_classifiers(), safe_k=3)
ds_test.processed_dsel = dsel_processed_ex1
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
ds_test.neighbors = np.array([0, 1, 2, 6, 7, 8, 14])
result = ds_test._frienemy_pruning()
assert np.array_equal(result, np.array([[1, 1, 1]]))
def test_frienemy_not_all_classifiers_crosses():
ds_test = DS(create_pool_classifiers(), safe_k=3)
ds_test.processed_dsel = dsel_processed_ex1
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
ds_test.neighbors = neighbors_ex1[0, :]
result = ds_test._frienemy_pruning()
assert np.array_equal(result, np.array([[1, 1, 0]]))
def test_predict_proba_IH_knn(index):
query = np.atleast_2d([1, 1])
ds_test = DS(create_pool_classifiers(), with_IH=True, IH_rate=0.5)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.dsel_scores = dsel_scores_ex1
ds_test.neighbors = neighbors_ex1[index, :]
ds_test.distances = distances_ex1[index, :]
ds_test.roc_algorithm.predict_proba = MagicMock(return_value=np.atleast_2d([0.45, 0.55]))
proba = ds_test.predict_proba(query)
assert np.isclose(proba, np.atleast_2d([0.45, 0.55])).all()
def test_label_encoder_only_dsel():
X_dsel_ex1 = np.array([[-1, 1], [-0.75, 0.5], [-1.5, 1.5]])
y_dsel_ex1 = np.array(['cat', 'dog', 'plane'])
query = np.atleast_2d([[1, 0], [-1, -1]])
ds_test = DS(create_pool_classifiers_dog_cat_plane(), k=2)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.neighbors = neighbors_ex1[0, :]
ds_test.distances = distances_ex1[0, :]
ds_test.classify_instance = Mock()
ds_test.classify_instance.side_effect = [1, 0]
predictions = ds_test.predict(query)
assert np.array_equal(predictions, ['dog', 'cat'])
def test_frienemy_not_all_classifiers_crosses_batch():
expected = np.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
ds_test = DS(create_pool_classifiers(), safe_k=3)
ds_test.processed_dsel = dsel_processed_ex1
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
# passing three samples to compute the DFP at the same time
ds_test.neighbors = neighbors_ex1
result = ds_test._frienemy_pruning()
assert np.array_equal(result, expected)
def test_input_X_1D():
X = np.ones(10)
ds_test = DS(create_pool_classifiers())
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
with pytest.raises(Warning):
ds_test.predict(X)
def test_frienemy_no_classifier_crosses():
X = X_dsel_ex1
y = y_dsel_ex1
ds_test = DS(create_pool_classifiers())
ds_test.fit(X, y)
ds_test.neighbors = neighbors_ex1[0, :]
mask = ds_test._frienemy_pruning()
assert mask.size == 3 and mask.all() == 1
def test_label_encoder_only_dsel():
X_dsel_ex1 = np.array([[-1, 1], [-0.75, 0.5], [-1.5, 1.5]])
y_dsel_ex1 = np.array(['cat', 'dog', 'plane'])
query = np.atleast_2d([[1, 0], [-1, -1]])
ds_test = DS(create_pool_classifiers_dog_cat_plane(), k=2)
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.neighbors = neighbors_ex1[0, :]
ds_test.distances = distances_ex1[0, :]
ds_test.classify_with_ds = Mock()
ds_test.classify_with_ds.return_value = [
1, 0
] # changed here due to batch processing
predictions = ds_test.predict(query)
assert np.array_equal(predictions, ['dog', 'cat'])
def test_frienemy_no_classifier_crosses():
X = X_dsel_ex1
y = y_dsel_ex1
ds_test = DS(create_pool_classifiers())
ds_test.fit(X, y)
ds_test.neighbors = neighbors_ex1[0, :]
mask = ds_test._frienemy_pruning()
assert mask.shape == (1, 3) and np.allclose(mask, 1)
def test_frienemy_safe_region_batch():
n_samples = 10
n_classifiers = 3
expected = np.ones((n_samples, n_classifiers))
ds_test = DS(create_pool_classifiers(), safe_k=3)
ds_test.processed_dsel = dsel_processed_ex1
ds_test.DSEL_target = y_dsel_ex1
ds_test.DSEL_data = X_dsel_ex1
ds_test.neighbors = np.tile(np.array([0, 1, 2, 6, 7, 8, 14]),
(n_samples, 1))
result = ds_test._frienemy_pruning()
assert np.array_equal(result, expected)
def test_predict_proba_all_agree():
query = np.atleast_2d([1, 1])
ds_test = DS(create_pool_classifiers())
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
ds_test.dsel_scores = dsel_scores_ex1
backup_all_agree = DS._all_classifier_agree
DS._all_classifier_agree = MagicMock(return_value=np.array([True]))
proba = ds_test.predict_proba(query)
DS._all_classifier_agree = backup_all_agree
assert np.allclose(proba, np.atleast_2d([0.61, 0.39]))
def test_not_fitted_ds():
query = np.array([[1.0, 1.0]])
ds_test = DS(create_pool_classifiers())
with pytest.raises(NotFittedError):
ds_test.predict(query)
def test_all_classifiers_agree():
# 10 classifiers that return 1
predictions = np.ones((1, 10))
assert np.all(DS._all_classifier_agree(predictions))
def test_empty_pool():
pool_classifiers = []
with pytest.raises(ValueError):
DS(pool_classifiers)
def test_different_input_shape():
query = np.array([[1.0, 1.0, 2.0]])
ds_test = DS(create_pool_classifiers())
ds_test.fit(X_dsel_ex1, y_dsel_ex1)
with pytest.raises(ValueError):
ds_test.predict(query)
def test_valid_safe_k(k, safe_k):
with pytest.raises(ValueError):
DS([create_base_classifier(1)], k=k, safe_k=safe_k)
def test_check_safe_k_value(safe_k):
pool_classifiers = create_pool_classifiers()
with pytest.raises(ValueError):
DS(pool_classifiers, safe_k=safe_k)
