def test_make_prediction_errors_if_asked_for_proba_without_predict_proba_method(
self, train_iris_dataset: Dataset):
with pytest.raises(
MLToolingError,
match="LinearRegression does not have a `predict_proba`"):
model = Model(LinearRegression())
model.train_estimator(train_iris_dataset)
model.make_prediction(train_iris_dataset, 5, proba=True)
def test_bayessearch_best_model_is_not_fitted_if_refit_is_not_true(
self, pipeline_logistic: Pipeline, train_iris_dataset: Dataset):
model = Model(pipeline_logistic)
model, results = model.bayesiansearch(
train_iris_dataset,
param_distributions={"estimator__penalty": ["l1", "l2"]},
refit=False,
)
with pytest.raises(MLToolingError,
match="You haven't fitted the estimator"):
model.make_prediction(data=train_iris_dataset, idx=1)
def test_sql_dataset_raises_exception_when_load_prediction_data_returns_empty(
self, boston_sqldataset: Type[SQLDataset], regression: Model
):
class FailingDataset(boston_sqldataset):
def load_prediction_data(self, *args, **kwargs):
return pd.DataFrame()
data = FailingDataset("sqlite:///", schema=None)
with pytest.raises(
DatasetError, match="An empty dataset was returned by load_prediction_data"
):
regression.make_prediction(data, 0)
def test_filedataset_raises_exception_when_load_prediction_data_is_empty(
self, regression: Model, boston_filedataset,
boston_csv: pathlib.Path):
class FailingDataset(boston_filedataset):
def load_prediction_data(self, *args, **kwargs):
return pd.DataFrame()
data = FailingDataset(boston_csv).create_train_test()
with pytest.raises(
DatasetError,
match="An empty dataset was returned by load_prediction_data"):
regression.make_prediction(data, 0)
def test_load_prediction_data_works_as_expected(self):
dataset = load_demo_dataset("iris")
dataset.create_train_test(stratify=True)
feature_pipeline = Pipeline([("scale", DFStandardScaler())])
model = Model(LogisticRegression(), feature_pipeline=feature_pipeline)
model.train_estimator(dataset)
result = model.make_prediction(dataset, 5)
expected = pd.DataFrame({"Prediction": [0]})
pd.testing.assert_frame_equal(result, expected, check_dtype=False)
def test_make_prediction_with_regression_sqldataset_works_as_expected(
self, boston_sqldataset, loaded_boston_db):
dataset = boston_sqldataset(loaded_boston_db, schema=None)
dataset.create_train_test(stratify=False)
model = Model(LinearRegression())
model.train_estimator(dataset)
result = model.make_prediction(dataset, 0)
assert result.shape == (1, 1)
assert result.columns.tolist() == ["Prediction"]
def test_make_prediction_with_classification_sqldataset_works_as_expected(
self, iris_sqldataset, loaded_iris_db):
dataset = iris_sqldataset(loaded_iris_db, schema=None)
dataset.create_train_test()
model = Model(LogisticRegression(solver="lbfgs"))
model.train_estimator(dataset)
result = model.make_prediction(dataset, 0, proba=True)
assert result.shape == (1, 2)
assert result.columns.tolist() == [
"Probability Class 0", "Probability Class 1"
]
def test_make_prediction_returns_prediction_if_threshold_is_specified(
self,
classifier: Model,
use_index: bool,
expected_index: int,
train_iris_dataset: Dataset,
):
results = classifier.make_prediction(train_iris_dataset,
5,
threshold=0.6,
use_index=use_index)
assert isinstance(results, pd.DataFrame)
assert 2 == results.ndim
assert np.all((results == 1) | (results == 0))
assert np.all(np.sum(results, axis=1) == 0)
assert results.index == pd.RangeIndex(start=expected_index,
stop=expected_index + 1,
step=1)
def test_make_prediction_returns_proba_if_proba_is_true(
self,
classifier: Model,
use_index: bool,
expected_index: int,
train_iris_dataset: Dataset,
):
results = classifier.make_prediction(train_iris_dataset,
5,
proba=True,
use_index=use_index)
assert isinstance(results, pd.DataFrame)
assert 2 == results.ndim
assert np.all((results <= 1) & (results >= 0))
assert np.all(np.sum(results, axis=1) == 1)
assert results.index == pd.RangeIndex(start=expected_index,
stop=expected_index + 1,
step=1)
def test_make_prediction_errors_when_model_is_not_fitted(
self, train_iris_dataset):
model = Model(LinearRegression())
with pytest.raises(MLToolingError,
match="You haven't fitted the estimator"):
model.make_prediction(train_iris_dataset, 5)
