def test_posterior_computation(mocker, classes, prior, cfm, first_class_posterior):
def mock_confusion_matrix(y, y_pred):
return np.array(cfm)
mocker.patch("sklego.meta.subjective_classifier.confusion_matrix", side_effect=mock_confusion_matrix)
mock_estimator = mocker.Mock(RandomForestClassifier())
mock_estimator.classes_ = np.array(classes)
subjective_model = SubjectiveClassifier(mock_estimator, dict(zip(classes, prior)))
subjective_model.fit(np.zeros((10, 10)), np.array([classes[0]] * 10))
assert (
pytest.approx(subjective_model.posterior_matrix_[0, 0], 0.001)
== first_class_posterior
)
assert np.isclose(subjective_model.posterior_matrix_.sum(axis=0), 1).all()
def test_fit_y_data_inconsistent_with_prior_failure_conditions(prior, y):
with pytest.raises(ValueError) as exc:
SubjectiveClassifier(RandomForestClassifier(),
prior).fit(np.zeros((len(y), 2)), np.array(y))
assert str(
exc.value).startswith('Training data is inconsistent with prior')
def test_params_failure_conditions(inner_estimator, prior, evidence,
expected_error_msg):
with pytest.raises(ValueError) as exc:
SubjectiveClassifier(inner_estimator, prior,
evidence).fit(np.zeros((2, 2)), np.zeros(2))
assert str(exc.value).startswith(expected_error_msg)
def test_weighted_proba(weights, y_hats, expected_probas):
assert np.isclose(
SubjectiveClassifier._weighted_proba(np.array(weights),
np.array(y_hats)),
np.array(expected_probas),
atol=1e-02,
).all()
def test_to_discrete():
assert np.isclose(
SubjectiveClassifier._to_discrete(
np.array([[1, 0], [0.8, 0.2], [0.5, 0.5], [0.2, 0.8]])
),
np.array([[1, 0], [1, 0], [1, 0], [0, 1]]),
).all()
def test_predict_proba(mocker, evidence_type, expected_probas):
def mock_confusion_matrix(y, y_pred):
return np.array([[80, 20], [10, 90]])
mocker.patch("sklego.meta.subjective_classifier.confusion_matrix", side_effect=mock_confusion_matrix)
mock_inner_estimator = mocker.Mock(RandomForestClassifier)
mock_inner_estimator.predict_proba.return_value = np.array(
[[0.8, 0.2], [1, 0], [0.5, 0.5], [0.2, 0.8]]
)
mock_inner_estimator.classes_ = np.array([0, 1])
subjective_model = SubjectiveClassifier(
mock_inner_estimator, {0: 0.8, 1: 0.2}, evidence=evidence_type
)
subjective_model.fit(np.zeros((10, 10)), np.zeros(10))
posterior_probabilities = subjective_model.predict_proba(np.zeros((4, 2)))
assert posterior_probabilities.shape == (4, 2)
assert np.isclose(
posterior_probabilities, np.array(expected_probas), atol=0.01
).all()
def test_estimator_checks_classification(test_fn):
if test_fn.__name__ == 'check_classifiers_classes':
prior = {
'one': 0.1,
'two': 0.1,
'three': 0.1,
-1: 0.1,
1: 0.6
} # nonsensical prior to make sklearn check pass
else:
prior = {0: 0.7, 1: 0.2, 2: 0.1}
# Some of the sklearn checkers generate random y data with 3 classes, so prior needs to have these classes
estimator = SubjectiveClassifier(LogisticRegression(), prior)
test_fn(SubjectiveClassifier.__name__, estimator)