def test_predict_value_error(random_X_data):
"""Raise ValueError if X doesn't have expected number of variables."""
X = create_random_X(random_X_data)
s = create_s()
lin_acf = counterfactually_fair_models.LinearACFClassifier()
lin_acf.fit(X, create_y(), s)
with pytest.raises(ValueError):
# pass in just a subset of the input variables
lin_acf.predict(X[:, 5], s)
lin_acf.predict_proba(X[:, 5], s)
def test_binary_single_class(random_X_data):
"""Linear ACF can handle training data with single-valued column."""
X = create_random_X(random_X_data)
X = np.concatenate([X, np.ones((X.shape[0], 1))], axis=1)
s = create_s()
y = create_y()
lin_acf = counterfactually_fair_models.LinearACFClassifier()
for residual_type in ["pearson", "deviance", "absolute"]:
lin_acf = counterfactually_fair_models.LinearACFClassifier(
binary_residual_type=residual_type)
lin_acf.fit(X, y, s)
lin_acf_pred_proba = lin_acf.predict_proba(X, s)[:, 1]
assert (
lin_acf.fit_residuals_ == lin_acf._compute_residuals_on_predict(
X, s)).all()
assert is_binary(lin_acf.predict(X, s))
assert is_continuous(lin_acf_pred_proba)
assert max(lin_acf_pred_proba) < 1
assert min(lin_acf_pred_proba) > 0
def test_fit_predict(random_X_data):
"""Test happy path of LinearACFClassifier `fit` and `predict` methods."""
X = create_random_X(random_X_data)
y = create_y()
s = create_s()
for residual_type in ["pearson", "deviance", "absolute"]:
lin_acf = counterfactually_fair_models.LinearACFClassifier(
binary_residual_type=residual_type)
lin_acf.fit(X, y, s)
lin_acf_pred_proba = lin_acf.predict_proba(X, s)[:, 1]
assert (
lin_acf.fit_residuals_ == lin_acf._compute_residuals_on_predict(
X, s)).all()
assert is_binary(lin_acf.predict(X, s))
assert is_continuous(lin_acf_pred_proba)
assert max(lin_acf_pred_proba) < 1
assert min(lin_acf_pred_proba) > 0
def test_invalid_binary_residual_type(random_X_data):
with pytest.raises(ValueError):
counterfactually_fair_models.LinearACFClassifier(
binary_residual_type="foobar")
def test_fairness_aware_meta_estimator():
X = create_linear_X()
y = create_y()
s = create_s()
# use sklearn estimator
lr = FairnessAwareMetaEstimator(LogisticRegression())
lr.fit(X, y)
lr.predict(X)
lr.predict_proba(X)
# use themis_ml LinearACFClassifier estimator
linear_acf = FairnessAwareMetaEstimator(
counterfactually_fair_models.LinearACFClassifier(
binary_residual_type="absolute"))
linear_acf.fit(X, y, s)
linear_acf.predict(X, s)
linear_acf.predict_proba(X, s)
# when fit/predict methods need s, raise ValueError
with pytest.raises(ValueError):
linear_acf.fit(X, y, s=None)
with pytest.raises(ValueError):
linear_acf.predict(X, s=None)
with pytest.raises(ValueError):
linear_acf.predict_proba(X, s=None)
# use themis_ml relabeller preprocessor
relabel_clf = FairnessAwareMetaEstimator(
estimator=LogisticRegression(), relabeller=relabelling.Relabeller())
relabel_clf.fit(X, y, s)
relabel_clf.predict(X)
relabel_clf.predict_proba(X)
# when estimator method does not need `s` on predict, raise ValueError if
# it is provided
with pytest.raises(ValueError):
relabel_clf.predict(X, s)
relabel_clf.predict_proba(X, s)
# use themis_ml RejectOption Classifier
single_reject_option_clf = FairnessAwareMetaEstimator(
reject_option_classification.SingleROClassifier())
single_reject_option_clf.fit(X, y)
single_reject_option_clf.predict(X, s)
single_reject_option_clf.predict_proba(X, s)
with pytest.raises(ValueError):
single_reject_option_clf.fit(X, y, s)
with pytest.raises(ValueError):
single_reject_option_clf.predict(X, s=None)
with pytest.raises(ValueError):
single_reject_option_clf.predict_proba(X, s=None)
multi_reject_option_clf = FairnessAwareMetaEstimator(
reject_option_classification.MultipleROClassifier())
multi_reject_option_clf.fit(X, y)
multi_reject_option_clf.predict(X, s)
multi_reject_option_clf.predict_proba(X, s)
with pytest.raises(ValueError):
multi_reject_option_clf.fit(X, y, s)
with pytest.raises(ValueError):
multi_reject_option_clf.predict(X, s=None)
with pytest.raises(ValueError):
multi_reject_option_clf.predict_proba(X, s=None)