Skip to content

GitLab

Unverified Commit 9afd8b93 authored by José Morales's avatar José Morales Committed by GitHub
Browse files

[tests][python-package] remove remaining tests using load_boston (fixes #4793) (#5581)

parent 6fa4673f
Hide whitespace changes
Inline Side-by-side
Showing with 62 additions and 77 deletions
tests/python_package_test/test_engine.py
View file @ 9afd8b93
......@@ -21,7 +21,7 @@ from sklearn.model_selection import GroupKFold, TimeSeriesSplit, train_test_spli
import lightgbm as lgb
from lightgbm.compat import PANDAS_INSTALLED, pd_DataFrame
from .utils import (SERIALIZERS, dummy_obj, load_boston, load_breast_cancer, load_digits, load_iris, logistic_sigmoid,
from .utils import (SERIALIZERS, dummy_obj, load_breast_cancer, load_digits, load_iris, logistic_sigmoid,
make_synthetic_regression, mse_obj, pickle_and_unpickle_object, sklearn_multiclass_custom_objective,
softmax)
......@@ -114,7 +114,8 @@ def test_rf():
@pytest.mark.parametrize('objective', ['regression', 'regression_l1', 'huber', 'fair', 'poisson'])
def test_regression(objective):
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
y = np.abs(y)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
params = {
'objective': objective,
......@@ -133,13 +134,13 @@ def test_regression(objective):
)
ret = mean_squared_error(y_test, gbm.predict(X_test))
if objective == 'huber':
assert ret < 35
assert ret < 430
elif objective == 'fair':
assert ret < 17
assert ret < 296
elif objective == 'poisson':
assert ret < 8
assert ret < 193
else:
assert ret < 7
assert ret < 338
assert evals_result['valid_0']['l2'][-1] == pytest.approx(ret)
......@@ -924,7 +925,7 @@ def test_early_stopping_min_delta(first_only, single_metric, greater_is_better):
def test_continue_train():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
params = {
'objective': 'regression',
......@@ -948,7 +949,7 @@ def test_continue_train():
init_model='model.txt'
)
ret = mean_absolute_error(y_test, gbm.predict(X_test))
assert ret < 2.0
assert ret < 13.6
assert evals_result['valid_0']['l1'][-1] == pytest.approx(ret)
np.testing.assert_allclose(evals_result['valid_0']['l1'], evals_result['valid_0']['custom_mae'])
......@@ -968,7 +969,7 @@ def test_continue_train_reused_dataset():
def test_continue_train_dart():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
params = {
'boosting_type': 'dart',
......@@ -989,7 +990,7 @@ def test_continue_train_dart():
init_model=init_gbm
)
ret = mean_absolute_error(y_test, gbm.predict(X_test))
assert ret < 2.0
assert ret < 13.6
assert evals_result['valid_0']['l1'][-1] == pytest.approx(ret)
......@@ -1920,10 +1921,12 @@ def test_refit_dataset_params():
np.testing.assert_allclose(stored_weights, refit_weight)
def test_mape_rf():
X, y = load_boston(return_X_y=True)
@pytest.mark.parametrize('boosting_type', ['rf', 'dart'])
def test_mape_for_specific_boosting_types(boosting_type):
X, y = make_synthetic_regression()
y = abs(y)
params = {
'boosting_type': 'rf',
'boosting_type': boosting_type,
'objective': 'mape',
'verbose': -1,
'bagging_freq': 1,
......@@ -1935,25 +1938,9 @@ def test_mape_rf():
gbm = lgb.train(params, lgb_train, num_boost_round=20)
pred = gbm.predict(X)
pred_mean = pred.mean()
assert pred_mean > 20
def test_mape_dart():
X, y = load_boston(return_X_y=True)
params = {
'boosting_type': 'dart',
'objective': 'mape',
'verbose': -1,
'bagging_freq': 1,
'bagging_fraction': 0.8,
'feature_fraction': 0.8,
'boost_from_average': False
}
lgb_train = lgb.Dataset(X, y)
gbm = lgb.train(params, lgb_train, num_boost_round=40)
pred = gbm.predict(X)
pred_mean = pred.mean()
assert pred_mean > 18
# the following checks that dart and rf with mape can predict outside the 0-1 range
# https://github.com/microsoft/LightGBM/issues/1579
assert pred_mean > 8
def check_constant_features(y_true, expected_pred, more_params):
......@@ -2667,19 +2654,22 @@ def test_model_size():
@pytest.mark.skipif(getenv('TASK', '') == 'cuda_exp', reason='Skip due to differences in implementation details of CUDA Experimental version')
def test_get_split_value_histogram():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X = np.repeat(X, 3, axis=0)
y = np.repeat(y, 3, axis=0)
X[:, 2] = np.random.default_rng(0).integers(0, 20, size=X.shape[0])
lgb_train = lgb.Dataset(X, y, categorical_feature=[2])
gbm = lgb.train({'verbose': -1}, lgb_train, num_boost_round=20)
# test XGBoost-style return value
params = {'feature': 0, 'xgboost_style': True}
assert gbm.get_split_value_histogram(**params).shape == (9, 2)
assert gbm.get_split_value_histogram(bins=999, **params).shape == (9, 2)
assert gbm.get_split_value_histogram(**params).shape == (12, 2)
assert gbm.get_split_value_histogram(bins=999, **params).shape == (12, 2)
assert gbm.get_split_value_histogram(bins=-1, **params).shape == (1, 2)
assert gbm.get_split_value_histogram(bins=0, **params).shape == (1, 2)
assert gbm.get_split_value_histogram(bins=1, **params).shape == (1, 2)
assert gbm.get_split_value_histogram(bins=2, **params).shape == (2, 2)
assert gbm.get_split_value_histogram(bins=6, **params).shape == (5, 2)
assert gbm.get_split_value_histogram(bins=7, **params).shape == (6, 2)
assert gbm.get_split_value_histogram(bins=6, **params).shape == (6, 2)
assert gbm.get_split_value_histogram(bins=7, **params).shape == (7, 2)
if lgb.compat.PANDAS_INSTALLED:
np.testing.assert_allclose(
gbm.get_split_value_histogram(0, xgboost_style=True).values,
......@@ -2700,8 +2690,8 @@ def test_get_split_value_histogram():
)
# test numpy-style return value
hist, bins = gbm.get_split_value_histogram(0)
assert len(hist) == 23
assert len(bins) == 24
assert len(hist) == 20
assert len(bins) == 21
hist, bins = gbm.get_split_value_histogram(0, bins=999)
assert len(hist) == 999
assert len(bins) == 1000
......@@ -2790,7 +2780,7 @@ def test_early_stopping_for_only_first_metric():
)
assert assumed_iteration == len(ret[list(ret.keys())[0]])
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
X_test1, X_test2, y_test1, y_test2 = train_test_split(X_test, y_test, test_size=0.5, random_state=73)
lgb_train = lgb.Dataset(X_train, y_train)
......@@ -2798,16 +2788,16 @@ def test_early_stopping_for_only_first_metric():
lgb_valid2 = lgb.Dataset(X_test2, y_test2, reference=lgb_train)
iter_valid1_l1 = 3
iter_valid1_l2 = 14
iter_valid2_l1 = 2
iter_valid1_l2 = 3
iter_valid2_l1 = 3
iter_valid2_l2 = 15
assert len(set([iter_valid1_l1, iter_valid1_l2, iter_valid2_l1, iter_valid2_l2])) == 4
assert len(set([iter_valid1_l1, iter_valid1_l2, iter_valid2_l1, iter_valid2_l2])) == 2
iter_min_l1 = min([iter_valid1_l1, iter_valid2_l1])
iter_min_l2 = min([iter_valid1_l2, iter_valid2_l2])
iter_min_valid1 = min([iter_valid1_l1, iter_valid1_l2])
iter_cv_l1 = 4
iter_cv_l2 = 12
iter_cv_l1 = 15
iter_cv_l2 = 13
assert len(set([iter_cv_l1, iter_cv_l2])) == 2
iter_cv_min = min([iter_cv_l1, iter_cv_l2])
......@@ -3153,7 +3143,7 @@ def test_trees_to_dataframe():
def test_interaction_constraints():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression(n_samples=200)
num_features = X.shape[1]
train_data = lgb.Dataset(X, label=y)
# check that constraint containing all features is equivalent to no constraint
......@@ -3166,9 +3156,7 @@ def test_interaction_constraints():
pred2 = est.predict(X)
np.testing.assert_allclose(pred1, pred2)
# check that constraint partitioning the features reduces train accuracy
est = lgb.train(dict(params, interaction_constraints=[list(range(num_features // 2)),
list(range(num_features // 2, num_features))]),
train_data, num_boost_round=10)
est = lgb.train(dict(params, interaction_constraints=[[0, 2], [1, 3]]), train_data, num_boost_round=10)
pred3 = est.predict(X)
assert mean_squared_error(y, pred1) < mean_squared_error(y, pred3)
# check that constraints consisting of single features reduce accuracy further
......@@ -3568,7 +3556,7 @@ def test_dump_model_hook():
@pytest.mark.skipif(getenv('TASK', '') == 'cuda_exp', reason='Forced splits are not yet supported by CUDA Experimental version')
def test_force_split_with_feature_fraction(tmp_path):
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
lgb_train = lgb.Dataset(X_train, y_train)
......@@ -3595,7 +3583,7 @@ def test_force_split_with_feature_fraction(tmp_path):
gbm = lgb.train(params, lgb_train)
ret = mean_absolute_error(y_test, gbm.predict(X_test))
assert ret < 2.0
assert ret < 15.7
tree_info = gbm.dump_model()["tree_info"]
assert len(tree_info) > 1
......
tests/python_package_test/test_sklearn.py
View file @ 9afd8b93
......@@ -21,8 +21,8 @@ from sklearn.utils.validation import check_is_fitted
import lightgbm as lgb
from lightgbm.compat import PANDAS_INSTALLED, pd_DataFrame
from .utils import (load_boston, load_breast_cancer, load_digits, load_iris, load_linnerud, make_ranking,
make_synthetic_regression, sklearn_multiclass_custom_objective, softmax)
from .utils import (load_breast_cancer, load_digits, load_iris, load_linnerud, make_ranking, make_synthetic_regression,
sklearn_multiclass_custom_objective, softmax)
decreasing_generator = itertools.count(0, -1)
task_to_model_factory = {
......@@ -112,12 +112,12 @@ def test_binary():
def test_regression():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
gbm = lgb.LGBMRegressor(n_estimators=50, verbose=-1)
gbm.fit(X_train, y_train, eval_set=[(X_test, y_test)], callbacks=[lgb.early_stopping(5)])
ret = mean_squared_error(y_test, gbm.predict(X_test))
assert ret < 7
assert ret < 174
assert gbm.evals_result_['valid_0']['l2'][gbm.best_iteration_ - 1] == pytest.approx(ret)
......@@ -226,12 +226,12 @@ def test_objective_aliases(custom_objective):
def test_regression_with_custom_objective():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
gbm = lgb.LGBMRegressor(n_estimators=50, verbose=-1, objective=objective_ls)
gbm.fit(X_train, y_train, eval_set=[(X_test, y_test)], callbacks=[lgb.early_stopping(5)])
ret = mean_squared_error(y_test, gbm.predict(X_test))
assert ret < 7.0
assert ret < 174
assert gbm.evals_result_['valid_0']['l2'][gbm.best_iteration_ - 1] == pytest.approx(ret)
......@@ -249,13 +249,12 @@ def test_binary_classification_with_custom_objective():
def test_dart():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
gbm = lgb.LGBMRegressor(boosting_type='dart', n_estimators=50)
gbm.fit(X_train, y_train)
score = gbm.score(X_test, y_test)
assert score >= 0.8
assert score <= 1.
assert 0.8 <= score <= 1.0
def test_stacking_classifier():
......@@ -280,7 +279,9 @@ def test_stacking_classifier():
def test_stacking_regressor():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression(n_samples=200)
n_features = X.shape[1]
n_input_models = 2
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)
regressors = [('gbm1', lgb.LGBMRegressor(n_estimators=3)),
('gbm2', lgb.LGBMRegressor(n_estimators=3))]
......@@ -291,11 +292,11 @@ def test_stacking_regressor():
score = reg.score(X_test, y_test)
assert score >= 0.2
assert score <= 1.
assert reg.n_features_in_ == 13 # number of input features
assert len(reg.named_estimators_['gbm1'].feature_importances_) == 13
assert reg.n_features_in_ == n_features # number of input features
assert len(reg.named_estimators_['gbm1'].feature_importances_) == n_features
assert reg.named_estimators_['gbm1'].n_features_in_ == reg.named_estimators_['gbm2'].n_features_in_
assert reg.final_estimator_.n_features_in_ == 15 # number of concatenated features
assert len(reg.final_estimator_.feature_importances_) == 15
assert reg.final_estimator_.n_features_in_ == n_features + n_input_models # number of concatenated features
assert len(reg.final_estimator_.feature_importances_) == n_features + n_input_models
def test_grid_search():
......@@ -765,7 +766,8 @@ def test_evaluate_train_set():
def test_metrics():
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression()
y = abs(y)
params = {'n_estimators': 2, 'verbose': -1}
params_fit = {'X': X, 'y': y, 'eval_set': (X, y)}
......@@ -1102,7 +1104,7 @@ def test_first_metric_only():
else:
assert gbm.n_estimators == gbm.best_iteration_
X, y = load_boston(return_X_y=True)
X, y = make_synthetic_regression(n_samples=300)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
X_test1, X_test2, y_test1, y_test2 = train_test_split(X_test, y_test, test_size=0.5, random_state=72)
params = {'n_estimators': 30,
......@@ -1114,11 +1116,11 @@ def test_first_metric_only():
params_fit = {'X': X_train,
'y': y_train}
iter_valid1_l1 = 3
iter_valid1_l2 = 18
iter_valid2_l1 = 11
iter_valid2_l2 = 7
assert len(set([iter_valid1_l1, iter_valid1_l2, iter_valid2_l1, iter_valid2_l2])) == 4
iter_valid1_l1 = 4
iter_valid1_l2 = 4
iter_valid2_l1 = 2
iter_valid2_l2 = 2
assert len(set([iter_valid1_l1, iter_valid1_l2, iter_valid2_l1, iter_valid2_l2])) == 2
iter_min_l1 = min([iter_valid1_l1, iter_valid2_l1])
iter_min_l2 = min([iter_valid1_l2, iter_valid2_l2])
iter_min = min([iter_min_l1, iter_min_l2])
......
tests/python_package_test/utils.py
View file @ 9afd8b93
......@@ -13,11 +13,6 @@ import lightgbm as lgb
SERIALIZERS = ["pickle", "joblib", "cloudpickle"]
@lru_cache(maxsize=None)
def load_boston(**kwargs):
return sklearn.datasets.load_boston(**kwargs)
@lru_cache(maxsize=None)
def load_breast_cancer(**kwargs):
return sklearn.datasets.load_breast_cancer(**kwargs)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment