def test_drift_adaptation_hatr():
dataset = synth.Friedman(seed=7).take(500)
model = tree.HoeffdingAdaptiveTreeRegressor(
leaf_prediction="model",
grace_period=50,
split_confidence=0.1,
adwin_confidence=0.1,
drift_window_threshold=10,
seed=7,
max_depth=3,
)
for i, (x, y) in enumerate(dataset):
y_ = y
if i > 250:
# Emulate an abrupt drift
y_ = 3 * y
model.learn_one(x, y_)
assert model._n_alternate_trees > 0
assert model._raw_memory_usage / (2 ** 20) < 0.025
@pytest.mark.parametrize(
"dataset, model",
[
(
get_regression_data(),
tree.HoeffdingTreeRegressor(
leaf_prediction="mean", max_size=0.5, memory_estimate_period=100
),
),
(
get_regression_data(),
tree.HoeffdingAdaptiveTreeRegressor(
leaf_prediction="mean", max_size=0.5, memory_estimate_period=100
),
),
],
)
def test_memory_usage_reg(dataset, model):
for x, y in dataset:
model.learn_one(x, y)
assert model._raw_memory_usage / (2 ** 20) < 0.5
def test_memory_usage_multilabel():
dataset = datasets.Music().take(500)
model = tree.LabelCombinationHoeffdingTreeClassifier(