def test_null():
print("Null model test")
print("Create custom null model")
(model, f_test) = regression_model_distribution(
H2OGradientBoostingEstimator,
upload_distribution(CustomDistributionGeneric, "null"))
print("rmse null model train:", model.rmse(valid=False))
print("rmse null model valid:", model.rmse(valid=True))
def test_worng_and_inherited_regression():
print("Create default gaussian model")
(model, f_test) = regression_model_default(H2OGradientBoostingEstimator,
"gaussian")
print("Create custom wrong gaussian model")
(model2, f_test2) = regression_model_distribution(
H2OGradientBoostingEstimator,
upload_distribution(CustomDistributionGaussianWrong, "gaussian_w"))
assert model.rmse(valid=False) != model2.rmse(valid=False), \
"Training rmse is not different for default and custom gaussian model."
assert model.rmse(valid=True) != model2.rmse(valid=True), \
"Validation rmse is not different for default and custom gaussian model."
print("Create custom gaussian model without inheritance")
(model3, f_test3) = regression_model_distribution(
H2OGradientBoostingEstimator,
upload_distribution(CustomDistributionGaussianNoInh, "gaussian_ni"))
check_model_metrics(model, model3, "gaussian_ni")
def test_regression():
print("Regression tests")
name = "gaussian"
print("Create default", name, "model")
(model, f_test) = regression_model_default(H2OGradientBoostingEstimator,
name)
print("Create custom ", name, "model")
(model2, f_test2) = regression_model_distribution(
H2OGradientBoostingEstimator,
upload_distribution(CustomDistributionGaussian, name))
check_model_metrics(model, model2, name)
def test_regression():
print("Regression tests")
name = "gaussian"
print("Create default", name, "model")
(model, f_test) = regression_model_default(H2OGradientBoostingEstimator,
name)
print("Create custom ", name, "model")
(model2, f_test2) = regression_model_distribution(
H2OGradientBoostingEstimator,
upload_distribution(CustomDistributionGaussian, name))
check_model_metrics(model, model2, name)
print(
"Test scoring history is without deviance if custom distribution is set."
)
sh = model.scoring_history().columns
shc = model2.scoring_history().columns
assert "training_deviance" in sh and "training_deviance" not in shc