def test_des_clustering():
pool_classifiers, X_dsel, y_dsel, X_test, y_test = setup_classifiers()
rng = np.random.RandomState(123456)
des_clustering = DESClustering(pool_classifiers, rng=rng)
des_clustering.fit(X_dsel, y_dsel)
assert np.isclose(des_clustering.score(X_test, y_test), 0.97872340425531912)
def test_diversity_metric_ratio():
test = DESClustering(create_pool_classifiers() * 10, metric='ratio')
# Mocking this method to avoid preprocessing the cluster information that
# is not required in this test.
test._preprocess_clusters = MagicMock(return_value=1)
test.fit(X_dsel_ex1, y_dsel_ex1)
assert test.diversity_func_ == ratio_errors
def test_classify_instance():
query = np.atleast_2d([1, -1])
clustering_test = DESClustering(create_pool_classifiers() * 4, k=2)
clustering_test.select = MagicMock(return_value=[0, 1, 2, 3, 5, 6, 7, 9])
predicted = clustering_test.classify_instance(query)
assert predicted == 0
def test_fit_heterogeneous_clusters(example_estimate_competence,
create_pool_classifiers):
clustering_test = DESClustering(create_pool_classifiers,
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5,
pct_diversity=0.33)
X, y = example_estimate_competence[0:2]
clustering_test.clustering.predict = MagicMock(
return_value=return_cluster_index_ex2)
clustering_test.fit(X, y)
# Index selected should be of any classifier that predicts the label 0
assert np.isclose(clustering_test.performance_cluster_[:, 1],
[0.428, 0.375],
atol=0.01).all()
assert np.isclose(clustering_test.performance_cluster_[:, 0],
[0.572, 0.625],
atol=0.01).all()
assert clustering_test.indices_[0,
0] == 0 or clustering_test.indices_[0,
0] == 2
assert clustering_test.indices_[1,
0] == 0 or clustering_test.indices_[1,
0] == 2
def test_classify_with_ds_single_sample():
query = np.ones(2)
predictions = np.array([0, 1, 0])
desknn_test = DESClustering(create_pool_classifiers())
desknn_test.select = MagicMock(return_value=np.array([[0, 2]]))
result = desknn_test.classify_with_ds(query, predictions)
assert np.allclose(result, 0)
def test_classify_with_ds_diff_sizes():
query = np.ones((10, 2))
predictions = np.ones((5, 3))
desknn_test = DESClustering(create_pool_classifiers())
with pytest.raises(ValueError):
desknn_test.classify_with_ds(query, predictions)
def test_classify_with_ds_diff_sizes():
query = np.ones((10, 2))
predictions = np.ones((5, 3))
des_clustering = DESClustering()
with pytest.raises(ValueError):
des_clustering.classify_with_ds(query, predictions)
def test_diversity_metric_DF(create_X_y):
X, y = create_X_y
test = DESClustering(metric='DF')
# Mocking this method to avoid preprocessing the cluster
# information that is not required in this test.
test._preprocess_clusters = MagicMock(return_value=1)
test.fit(X, y)
assert test.diversity_func_ == negative_double_fault
def test_des_clustering_proba():
pool_classifiers, X_dsel, y_dsel, X_test, y_test = setup_classifiers()
rng = np.random.RandomState(123456)
des_clustering = DESClustering(pool_classifiers, rng=rng)
des_clustering.fit(X_dsel, y_dsel)
probas = des_clustering.predict_proba(X_test)
expected = np.load('deslib/tests/expected_values/des_clustering_proba_integration.npy')
assert np.allclose(probas, expected)
def test_proba_with_ds_diff_sizes():
query = np.ones((10, 2))
predictions = np.ones((5, 3))
probabilities = np.ones((5, 3, 2))
des_clustering = DESClustering()
with pytest.raises(ValueError):
des_clustering.predict_proba_with_ds(query, predictions, probabilities)
def test_des_clustering():
from sklearn.cluster import KMeans
pool_classifiers, X_dsel, y_dsel, X_test, y_test = setup_classifiers()
rng = np.random.RandomState(123456)
cluster = KMeans(n_clusters=5, random_state=rng)
des_clustering = DESClustering(pool_classifiers, clustering=cluster)
des_clustering.fit(X_dsel, y_dsel)
assert np.isclose(des_clustering.score(X_test, y_test),
0.97872340425531912)
def test_select():
query = np.atleast_2d([1, -1])
clustering_test = DESClustering(create_pool_classifiers() * 2, k=2)
clustering_test.roc_algorithm.predict = MagicMock(return_value=[0])
clustering_test.indices = np.array([[0, 2], [1, 4]])
assert np.array_equal(clustering_test.select(query), [0, 2])
clustering_test.roc_algorithm.predict = MagicMock(return_value=[1])
assert np.array_equal(clustering_test.select(query), [1, 4])
def test_diversity_metric_ratio(create_X_y):
X, y = create_X_y
test = DESClustering(metric_diversity='ratio')
# Mocking this method to avoid preprocessing the cluster
# information that is not required in this test.
test._preprocess_clusters = MagicMock(return_value=1)
test.fit(X, y)
assert test.diversity_func_ == ratio_errors
def test_classify_instance():
query = np.atleast_2d([1, -1])
clustering_test = DESClustering(create_pool_classifiers() * 4, k=2)
clustering_test.select = MagicMock(return_value=[0, 1, 2, 3, 5, 6, 7, 9])
predictions = []
for clf in clustering_test.pool_classifiers:
predictions.append(clf.predict(query)[0])
predicted = clustering_test.classify_instance(query, np.array(predictions))
assert predicted == 0
def test_des_clustering_proba():
from sklearn.cluster import KMeans
pool_classifiers, X_dsel, y_dsel, X_test, y_test = setup_classifiers()
rng = np.random.RandomState(123456)
cluster = KMeans(n_clusters=5, random_state=rng)
des_clustering = DESClustering(pool_classifiers, clustering=cluster)
des_clustering.fit(X_dsel, y_dsel)
probas = des_clustering.predict_proba(X_test)
expected = np.load(
'deslib/tests/expected_values/des_clustering_proba_integration.npy')
assert np.allclose(probas, expected)
def test_classify_instance(create_pool_classifiers):
query = np.ones((1, 2))
clustering_test = DESClustering(create_pool_classifiers * 4,
clustering=KMeans(n_clusters=2))
clustering_test.select = MagicMock(return_value=[0, 1, 2, 3, 5, 6, 7, 9])
predictions = []
for clf in clustering_test.pool_classifiers:
predictions.append(clf.predict(query)[0])
predicted = clustering_test.classify_with_ds(query, np.array(predictions))
assert predicted == 0
def test_fit_homogeneous_clusters():
clustering_test = DESClustering(create_pool_classifiers()*2, k=2, pct_accuracy=0.5, pct_diversity=0.33)
clustering_test.roc_algorithm.fit_predict = MagicMock(return_value=return_cluster_index_ex1)
clustering_test.DFP_mask = np.ones(clustering_test.n_classifiers)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
assert clustering_test.accuracy_cluster[0, 1] == 0.0 and clustering_test.accuracy_cluster[0, [0, 2]].all() == 1.0
assert clustering_test.accuracy_cluster[1, 1] == 1.0 and clustering_test.accuracy_cluster[1, [0, 2]].all() == 0.0
for idx in clustering_test.indices[0, :]:
assert idx in (0, 2, 3, 5)
def test_fit_heterogeneous_clusters():
clustering_test = DESClustering(create_pool_classifiers(), k=2, pct_accuracy=0.5, pct_diversity=0.33)
clustering_test.roc_algorithm.fit_predict = MagicMock(return_value=return_cluster_index_ex2)
clustering_test.DFP_mask = np.ones(clustering_test.n_classifiers)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
# Index selected should be of any classifier that predicts the class label 0
assert np.isclose(clustering_test.accuracy_cluster[:, 1], [0.428, 0.375], atol=0.01).all()
assert np.isclose(clustering_test.accuracy_cluster[:, 0], [0.572, 0.625], atol=0.01).all()
assert clustering_test.indices[0, 0] == 0 or clustering_test.indices[0, 0] == 2
assert clustering_test.indices[1, 0] == 0 or clustering_test.indices[1, 0] == 2
def test_fit_clusters_less_diverse(example_estimate_competence,
create_pool_classifiers):
clustering_test = DESClustering(create_pool_classifiers * 2,
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5, pct_diversity=0.33,
more_diverse=False)
X, y = example_estimate_competence[0:2]
clustering_test.clustering.predict = MagicMock(return_value=y)
clustering_test.fit(X, y)
assert (clustering_test.performance_cluster_[0, 1] == 0.0 and
clustering_test.performance_cluster_[0, [0, 2]].all() == 1.0)
assert (clustering_test.performance_cluster_[1, 1] == 1.0 and
clustering_test.performance_cluster_[1, [0, 2]].all() == 0.0)
for idx in clustering_test.indices_[0, :]:
assert idx in (1, 3, 4, 5)
def test_fit_clusters_less_diverse():
clustering_test = DESClustering(create_pool_classifiers() * 2,
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5,
pct_diversity=0.33,
more_diverse=False)
clustering_test.clustering.predict = MagicMock(
return_value=return_cluster_index_ex1)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
assert (clustering_test.accuracy_cluster_[0, 1] == 0.0
and clustering_test.accuracy_cluster_[0, [0, 2]].all() == 1.0)
assert (clustering_test.accuracy_cluster_[1, 1] == 1.0
and clustering_test.accuracy_cluster_[1, [0, 2]].all() == 0.0)
for idx in clustering_test.indices_[0, :]:
assert idx in (1, 3, 4, 5)
def test_fit_homogeneous_clusters(create_pool_classifiers,
example_estimate_competence):
clustering_test = DESClustering(create_pool_classifiers * 2,
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5,
pct_diversity=0.33)
X, y = example_estimate_competence[0:2]
clustering_test.clustering.predict = MagicMock(return_value=y)
clustering_test.fit(X, y)
assert (clustering_test.accuracy_cluster_[0, 1] == 0.0
and clustering_test.accuracy_cluster_[0, [0, 2]].all() == 1.0)
assert (clustering_test.accuracy_cluster_[1, 1] == 1.0
and clustering_test.accuracy_cluster_[1, [0, 2]].all() == 0.0)
for idx in clustering_test.indices_[0, :]:
assert idx in (0, 2, 3, 5)
def test_fit_homogeneous_clusters():
clustering_test = DESClustering(create_pool_classifiers() * 2,
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5,
pct_diversity=0.33)
clustering_test.clustering.predict = MagicMock(
return_value=return_cluster_index_ex1)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
assert clustering_test.accuracy_cluster_[
0, 1] == 0.0 and clustering_test.accuracy_cluster_[0,
[0, 2]].all() == 1.0
assert clustering_test.accuracy_cluster_[
1, 1] == 1.0 and clustering_test.accuracy_cluster_[1,
[0, 2]].all() == 0.0
for idx in clustering_test.indices_[0, :]:
assert idx in (0, 2, 3, 5)
def test_fit_heterogeneous_clusters():
clustering_test = DESClustering(create_pool_classifiers(),
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5,
pct_diversity=0.33)
clustering_test.clustering.predict = MagicMock(
return_value=return_cluster_index_ex2)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
# Index selected should be of any classifier that predicts the label 0
assert np.isclose(clustering_test.accuracy_cluster_[:, 1], [0.428, 0.375],
atol=0.01).all()
assert np.isclose(clustering_test.accuracy_cluster_[:, 0], [0.572, 0.625],
atol=0.01).all()
assert clustering_test.indices_[0,
0] == 0 or clustering_test.indices_[0,
0] == 2
assert clustering_test.indices_[1,
0] == 0 or clustering_test.indices_[1,
0] == 2
def test_fit_clusters_less_diverse():
clustering_test = DESClustering(create_pool_classifiers() * 2,
k=2,
pct_accuracy=1.0,
pct_diversity=0.60,
more_diverse=False)
clustering_test.roc_algorithm.fit_predict = MagicMock(
return_value=return_cluster_index_ex1)
clustering_test.DFP_mask = np.ones(clustering_test.n_classifiers)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
assert clustering_test.accuracy_cluster[
0, 1] == 0.0 and clustering_test.accuracy_cluster[0,
[0, 2]].all() == 1.0
assert clustering_test.accuracy_cluster[
1, 1] == 1.0 and clustering_test.accuracy_cluster[1,
[0, 2]].all() == 0.0
assert np.isin(clustering_test.indices[0, :], np.array([1, 3, 5, 4])).all()
def test_estimate_competence(create_pool_classifiers,
example_estimate_competence):
query = np.atleast_2d([1, 1])
clustering_test = DESClustering(create_pool_classifiers * 2,
clustering=KMeans(n_clusters=2),
pct_accuracy=0.5,
pct_diversity=0.33)
X, y = example_estimate_competence[0:2]
# Keep the original predict method to change after
clustering_test.clustering.predict = MagicMock(
return_value=return_cluster_index_ex2)
clustering_test.fit(X, y)
clustering_test.clustering_.predict = MagicMock(return_value=0)
competences = clustering_test.estimate_competence(query)
assert np.array_equal(competences,
clustering_test.performance_cluster_[0, :])
clustering_test.clustering_.predict = MagicMock(return_value=1)
competences = clustering_test.estimate_competence(query)
assert np.array_equal(competences,
clustering_test.performance_cluster_[1, :])
def test_select():
query = np.atleast_2d([1, -1])
clustering_test = DESClustering()
clustering_test.clustering_ = KMeans()
clustering_test.clustering_.predict = MagicMock(return_value=[0])
clustering_test.indices_ = np.array([[0, 2], [1, 4]])
assert np.array_equal(clustering_test.select(query), [[0, 2]])
clustering_test.clustering_.predict = MagicMock(return_value=[1])
assert np.array_equal(clustering_test.select(query), [[1, 4]])
def test_estimate_competence():
query = np.atleast_2d([1, 1])
clustering_test = DESClustering(create_pool_classifiers(), k=2, pct_accuracy=0.5, pct_diversity=0.33)
clustering_test.roc_algorithm.fit_predict = MagicMock(return_value=return_cluster_index_ex2)
clustering_test.DFP_mask = np.ones(clustering_test.n_classifiers)
clustering_test.fit(X_dsel_ex1, y_dsel_ex1)
clustering_test.roc_algorithm.predict = MagicMock(return_value=0)
competences = clustering_test.estimate_competence(query)
assert np.array_equal(competences, clustering_test.accuracy_cluster[0, :])
clustering_test.roc_algorithm.predict = MagicMock(return_value=1)
competences = clustering_test.estimate_competence(query)
assert np.array_equal(competences, clustering_test.accuracy_cluster[1, :])
def test_diversity_metric_ratio():
test = DESClustering(create_pool_classifiers() * 10, metric='ratio')
assert test.diversity_func == ratio_errors
def test_diversity_metric_DF():
test = DESClustering(create_pool_classifiers() * 10, metric='DF')
assert test.diversity_func == negative_double_fault
def test_diversity_metric_Q():
test = DESClustering(create_pool_classifiers() * 10, metric='Q')
assert test.diversity_func == Q_statistic