def test_grouped_returns_numpy(pipeline, groupby, is_estimator, input_df):
gp = GroupedPipeline(groupby=groupby, pipeline=pipeline)
if is_estimator:
out = gp.fit_predict(input_df)
else:
out = gp.fit_transform(input_df)
assert type(out) is np.ndarray
def test_grouped_order(pipeline, groupby, is_estimator, input_df):
gp = GroupedPipeline(groupby=groupby, pipeline=pipeline)
if is_estimator:
out = gp.fit_predict(input_df).values
expected = input_df[ini.Columns.target].values
else:
out = gp.fit_transform(input_df).values
expected = input_df[[ini.Columns.target]].values
np.testing.assert_array_equal(out, expected)
def test_iter_groups_non_consecutive_index(index):
group = [1] * 2 + [2] * (len(index) - 2)
value = np.random.random(len(index))
input_df = pd.DataFrame(
[group, value], index=["group", "value"], columns=index
).T
gp = GroupedPipeline(groupby=["group"], pipeline=None)
for key, sub_df, _ in gp._iter_groups(input_df):
assert (sub_df["group"] == key).all()
def test_grouped_with_y(pipeline, groupby, y, input_df, y_mode):
if y_mode == 'series':
y = input_df[y]
elif y_mode == 'array':
y = input_df[y].values
gp = GroupedPipeline(groupby=groupby, pipeline=pipeline)
out = gp.fit_predict(input_df, y)
assert type(out) is np.ndarray
assert len(out) == len(input_df)
def test_one_group_missing_return_none(input_df):
transform_df = mock.mock_raw_data(ids=[0, 1, 2])
gp = GroupedPipeline(groupby=['id'], pipeline=Pipeline([val_selector]),
errors='return_empty')
gp.fit(input_df)
out = gp.transform(transform_df)
assert out.shape[1] == 1
transformed_part = transform_df[ini.Columns.target].values[:96]
np.testing.assert_array_equal(out[:96, 0], transformed_part)
assert np.isnan(out[96:]).all()
def test_grouped_values(mocker, pipeline, groupby, is_estimator, input_df):
gp = GroupedPipeline(groupby=groupby, pipeline=pipeline)
if is_estimator:
out = gp.fit_predict(input_df)
else:
out = gp.fit_transform(input_df)
input_df['out'] = out
for _, df in input_df.groupby(groupby):
expected = df[ini.Columns.target].shift(1)
np.testing.assert_array_equal(expected.values, df['out'].values)
def test_one_groups_missing_return_df(input_df):
transform_df = mock.mock_raw_data(ids=[0, 1, 2])
dt_feat = PandasDateTimeFeaturizer(attributes='month')
gp = GroupedPipeline(groupby=['id'], pipeline=dt_feat, errors='return_df')
gp.fit(input_df)
out = gp.transform(transform_df)
set(out.columns) == {
'id', ini.Columns.datetime, ini.Columns.target, 'month'
}
assert (~out[out.id == 0].month.isnull()).all()
assert (~out[out.id == 1].month.isnull()).all()
assert (out[out.id == 2].month.isnull()).all()
orig_cols = ['id', ini.Columns.datetime, ini.Columns.target]
pd.testing.assert_frame_equal(out[orig_cols], transform_df)
def test_all_groups_missing_raises(input_df, errors):
transform_df = mock.mock_raw_data(ids=[2, 3])
gp = GroupedPipeline(groupby=['id'], pipeline=Pipeline([col_selector]),
errors=errors)
gp.fit(input_df)
with pytest.raises(KeyError,
message='All keys missing in fitted pipelines'):
gp.transform(transform_df)
test_pipeline1 = Pipeline([('select',
PandasDateTimeFeaturizer(column='datetime',
attributes='hour')),
('select2', PandasValueSelector(['hour', 'id']))])
test_pipeline2 = Pipeline([('select',
PandasDateTimeFeaturizer(column='datetime',
attributes='hour')),
('select2', PandasValueSelector(['id']))])
test_pipeline3 = Pipeline([('select',
PandasDateTimeFeaturizer(column='datetime',
attributes='hour')),
('select2', PandasColumnSelector(['hour', 'id']))])
test_grouped_pipeline1 = GroupedPipeline(groupby='team',
pipeline=test_pipeline1)
test_grouped_pipeline1_multiple = GroupedPipeline(groupby=['week', 'team'],
pipeline=test_pipeline1)
test_grouped_pipeline2 = GroupedPipeline(groupby='team',
pipeline=test_pipeline2)
test_grouped_pipeline3 = GroupedPipeline(groupby='team',
pipeline=test_pipeline3)
# feature union
pipeline1 = Pipeline([('featurize',
PandasDateTimeFeaturizer(column='datetime',
attributes='hour')),
('select', PandasColumnSelector(columns=['hour']))])
pipeline2 = Pipeline([('select2', PandasColumnSelector('id'))])
pipeline3 = Pipeline([('select3', PandasValueSelector('id'))])
pipeline4 = Pipeline([('featurize',
def test_grouped_returns_pandas(pipeline, groupby, input_df):
gp = GroupedPipeline(groupby=groupby, pipeline=pipeline)
out = gp.fit_transform(input_df)
assert type(out) is pd.DataFrame
def test_raises_when_missing_key(input_df):
transform_df = mock.mock_raw_data(ids=[0, 1, 2])
gp = GroupedPipeline(groupby=['id'], pipeline=Pipeline([col_selector]))
gp.fit(input_df)
with pytest.raises(KeyError, message="Missing key 2 in fitted pipelines"):
gp.transform(transform_df)
