X, y = make_classification(random_state=0)
y[:] = 1 # constant target will lead to a single root node
est = Est(max_iter=20)
est.fit(X, y)
assert all(len(predictor[0].nodes) == 1 for predictor in est._predictors)
assert all(predictor[0].nodes[0]['value'] == 0
for predictor in est._predictors)
# Still gives correct predictions thanks to the baseline prediction
assert_allclose(est.predict(X), y)
@pytest.mark.parametrize(
'Est, loss, X, y',
[(HistGradientBoostingClassifier, BinaryCrossEntropy(sample_weight=None),
X_classification, y_classification),
(HistGradientBoostingRegressor, LeastSquares(sample_weight=None),
X_regression, y_regression)])
def test_custom_loss(Est, loss, X, y):
est = Est(loss=loss, max_iter=20)
est.fit(X, y)
@pytest.mark.parametrize('HistGradientBoosting, X, y', [
(HistGradientBoostingClassifier, X_classification, y_classification),
(HistGradientBoostingRegressor, X_regression, y_regression),
(HistGradientBoostingClassifier, X_multi_classification,
y_multi_classification),
])
def test_staged_predict(HistGradientBoosting, X, y):
est = Est(max_iter=20)
est.fit(X, y)
assert all(len(predictor[0].nodes) == 1 for predictor in est._predictors)
assert all(predictor[0].nodes[0]["value"] == 0
for predictor in est._predictors)
# Still gives correct predictions thanks to the baseline prediction
assert_allclose(est.predict(X), y)
@pytest.mark.parametrize(
"Est, loss, X, y",
[
(
HistGradientBoostingClassifier,
BinaryCrossEntropy(sample_weight=None),
X_classification,
y_classification,
),
(
HistGradientBoostingRegressor,
LeastSquares(sample_weight=None),
X_regression,
y_regression,
),
],
)
def test_custom_loss(Est, loss, X, y):
est = Est(loss=loss, max_iter=20)
est.fit(X, y)