def test_sklearn_regression():
# supports weights
check_regression(SklearnRegressor(clf=AdaBoostRegressor(n_estimators=50)))
# doesn't support weights
check_regression(
SklearnRegressor(clf=GradientBoostingRegressor(n_estimators=50)),
supports_weight=False)
def test_own_regression_reports():
"""
testing regressor.test_on
"""
X, y, sample_weight = generate_regression_data()
regressor = SklearnRegressor(RandomForestRegressor())
regressor.fit(X, y, sample_weight=sample_weight)
report = regressor.test_on(X, y, sample_weight=sample_weight)
mse1 = report.compute_metric(mean_squared_error)
lds = LabeledDataStorage(X, y, sample_weight=sample_weight)
mse2 = regressor.test_on_lds(lds=lds).compute_metric(mean_squared_error)
assert mse1 == mse2, 'Something wrong with test_on'
def test_gridsearch_sklearn_regression():
scorer = RegressionFoldingScorer(mean_squared_error)
grid_param = OrderedDict({
"n_estimators": [10, 20],
"learning_rate": [0.1, 0.05],
'features': [['column0', 'column1'], ['column0', 'column1', 'column2']]
})
generator = RegressionParameterOptimizer(grid_param, n_evaluations=4)
grid = GridOptimalSearchCV(SklearnRegressor(clf=AdaBoostRegressor()),
generator, scorer)
# parallel_profile='threads-3')
_ = check_grid(grid,
False,
False,
False,
use_weights=True,
classification=False)
regressor = check_grid(grid,
False,
False,
False,
use_weights=False,
classification=False)
# Check parameters of best fitted classifier
assert 2 <= len(regressor.features) <= 3, 'Features were not set'
params = regressor.get_params()
for key in grid_param:
if key in params:
assert params[key] == grid.generator.best_params_[key]
else:
assert params['clf__' + key] == grid.generator.best_params_[key]
def test_folding_regressor(n_samples=100, n_features=3):
"""
checking mostly different things with quality of predictions and not using train data during predictions.
"""
from sklearn.metrics import mean_squared_error
X = numpy.random.normal(size=[n_samples, n_features])
y = numpy.random.normal(size=n_samples)
kfolder = FoldingRegressor(SklearnRegressor(GradientBoostingRegressor()),
n_folds=2)
kfolder.fit(X, y)
preds = kfolder.predict(X)
# checking that we fitted fine
assert mean_squared_error(y,
preds) > mean_squared_error(y * 0., preds) * 0.5
# shuffled predictions
p = numpy.random.permutation(n_samples)
preds2 = kfolder.predict(X[p])[numpy.argsort(p)]
# Now let's compare this with shuffled kFolding:
assert mean_squared_error(y, preds) > mean_squared_error(y, preds2) * 0.5
preds_mean = kfolder.predict(X,
vote_function=lambda x: numpy.mean(x, axis=0))
# Now let's compare this with mean prediction:
assert mean_squared_error(y, preds) > mean_squared_error(y, preds_mean)
def test_folding_regressor_functions():
"""Testing folding functions """
data, y, sample_weight = generate_classification_data()
for X in [data, numpy.array(data)]:
kfolder = FoldingRegressor(SklearnRegressor(GradientBoostingRegressor(n_estimators=5)), n_folds=2)
kfolder.fit(X, y, sample_weight=sample_weight)
preds = kfolder.predict(X)
for p in kfolder.staged_predict(X):
pass
assert numpy.allclose(p, preds)
importances = kfolder.feature_importances_
other_importances = kfolder.get_feature_importances()
def test_own_regression_reports():
"""
testing regressor.test_on
"""
X, y, sample_weight = generate_regression_data()
regressor = SklearnRegressor(RandomForestRegressor())
regressor.fit(X, y, sample_weight=sample_weight)
report = regressor.test_on(X, y, sample_weight=sample_weight)
mse1 = report.compute_metric(mean_squared_error)
lds = LabeledDataStorage(X, y, sample_weight=sample_weight)
mse2 = regressor.test_on_lds(lds=lds).compute_metric(mean_squared_error)
assert mse1 == mse2, 'Something wrong with test_on'
def test_folding_regressor_with_check_model():
base_clf = SklearnRegressor(GradientBoostingRegressor(n_estimators=4))
folding_str = FoldingRegressor(base_clf, n_folds=2)
check_regression(folding_str, True, True, True)
def test_sklearn_regression():
check_regression(SklearnRegressor(clf=AdaBoostRegressor(n_estimators=50)))
check_regression(
SklearnRegressor(clf=GradientBoostingRegressor(n_estimators=50)))