def test_evaluate_only_added_features_true(self):
"""
The boolean flag `evaluate_only_extracted_features` makes sure that only the time series based features are
filtered. This unit tests checks that
"""
augmenter = RelevantFeatureAugmenter(
feature_extraction_settings=self.extraction_settings,
evaluate_only_added_features=True,
column_value="val",
column_id="id",
column_sort="sort",
column_kind="kind")
y = pd.Series({10: 1, 500: 0})
X = pd.DataFrame(index=[10, 500])
X["pre_feature"] = 0
augmenter.set_timeseries_container(self.test_df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(
sum(["pre_feature" == column for column in transformed_X.columns]),
1)
def test_evaluate_only_added_features_false(self):
"""
The boolean flag `evaluate_only_extracted_features` makes sure that only the time series based features are
filtered. This unit tests checks that
"""
augmenter = RelevantFeatureAugmenter(
feature_extraction_settings=self.extraction_settings,
evaluate_only_added_features=False,
column_value="val",
column_id="id",
column_sort="sort",
column_kind="kind")
df, y = self.create_test_data_sample_with_target()
X = pd.DataFrame(index=np.unique(df.id))
X["pre_drop"] = 0
X["pre_keep"] = y
augmenter.set_timeseries_container(df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(
sum(["pre_keep" == column for column in transformed_X.columns]), 1)
self.assertEqual(
sum(["pre_drop" == column for column in transformed_X.columns]), 0)
def test_no_ids_present(self):
augmenter = RelevantFeatureAugmenter(kind_to_fc_parameters=self.kind_to_fc_parameters,
filter_only_tsfresh_features=False,
column_value="val", column_id="id", column_sort="sort", column_kind="kind")
df, y = self.create_test_data_sample_with_target()
X_with_wrong_ids = pd.DataFrame(index=[-999])
augmenter.set_timeseries_container(df)
self.assertRaisesRegex(AttributeError, r"The ids of the time series container",
augmenter.fit, X_with_wrong_ids, y)
def test_does_impute(self, calculate_relevance_table_mock):
df = pd.DataFrame([[1, 1, 1], [2, 1, 1]], columns=['id', 'time', 'value'])
X = pd.DataFrame(index=[1])
y = pd.Series([0, 1])
fc_parameters = {"autocorrelation": [{'lag': 2}]}
calculate_relevance_table_mock.return_value = pd.DataFrame(columns=['feature', 'p_value', 'relevant'])
augmenter = RelevantFeatureAugmenter(column_id='id', column_sort='time', default_fc_parameters=fc_parameters)
augmenter.set_timeseries_container(df)
augmenter.fit(X, y)
assert calculate_relevance_table_mock.call_count == 1
assert not calculate_relevance_table_mock.call_args[0][0].isnull().any().any()
def test_does_impute(self, calculate_relevance_table_mock):
df = pd.DataFrame([[1, 1, 1], [2, 1, 1]], columns=['id', 'time', 'value'])
X = pd.DataFrame(index=[1])
y = pd.Series([0, 1])
fc_parameters = {"autocorrelation": [{'lag': 2}]}
calculate_relevance_table_mock.return_value = pd.DataFrame(columns=['feature', 'p_value', 'relevant'])
augmenter = RelevantFeatureAugmenter(column_id='id', column_sort='time', default_fc_parameters=fc_parameters)
augmenter.set_timeseries_container(df)
augmenter.fit(X, y)
assert calculate_relevance_table_mock.call_count == 1
assert not calculate_relevance_table_mock.call_args[0][0].isnull().any().any()
def test_nothing_relevant(self):
augmenter = RelevantFeatureAugmenter(kind_to_fc_parameters=self.kind_to_fc_parameters,
column_value="val", column_id="id", column_sort="sort",
column_kind="kind")
y = pd.Series({10: 1, 500: 0})
X = pd.DataFrame(index=[10, 500])
augmenter.set_timeseries_container(self.test_df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(list(transformed_X.columns), [])
self.assertEqual(list(transformed_X.index), list(X.index))
def test_nothing_relevant(self):
augmenter = RelevantFeatureAugmenter(kind_to_fc_parameters=self.kind_to_fc_parameters,
column_value="val", column_id="id", column_sort="sort",
column_kind="kind")
y = pd.Series({10: 1, 500: 0})
X = pd.DataFrame(index=[10, 500])
augmenter.set_timeseries_container(self.test_df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(list(transformed_X.columns), [])
self.assertEqual(list(transformed_X.index), list(X.index))
def test_multiclass_selection(self):
augmenter = RelevantFeatureAugmenter(
column_value="val",
column_id="id",
column_sort="sort",
column_kind="kind",
multiclass=True,
n_significant=3,
)
df, y = self.create_test_data_sample_with_multiclass_target()
X = pd.DataFrame(index=np.unique(df.id))
augmenter.set_timeseries_container(df)
fit_transformed_X = augmenter.fit_transform(X, y)
self.assertEqual(len(fit_transformed_X.columns), 4)
def test_impute_works(self):
self.extraction_settings.kind_to_calculation_settings_mapping["a"].update({"kurtosis": None})
augmeter = RelevantFeatureAugmenter(feature_extraction_settings=self.extraction_settings,
column_value="val", column_id="id", column_sort="sort",
column_kind="kind")
y = pd.Series({10: 1, 500: 0})
X = pd.DataFrame(index=[10, 500])
augmeter.set_timeseries_container(self.test_df)
augmeter.fit(X, y)
transformed_X = augmeter.transform(X.copy())
self.assertEqual(list(transformed_X.columns), [])
self.assertEqual(list(transformed_X.index), list(X.index))
def test_nothing_relevant(self):
augmenter = RelevantFeatureAugmenter(
feature_extraction_settings=self.extraction_settings,
column_value="val",
column_id="id",
column_sort="sort",
column_kind="kind")
y = pd.Series({1: 1, 5: 0})
X = pd.DataFrame(index=[1, 5])
augmenter.set_timeseries_container(self.test_df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(list(transformed_X.columns), [])
self.assertEqual(list(transformed_X.index), list(X.index))
def test_evaluate_only_added_features_true(self):
"""
The boolean flag `evaluate_only_extracted_features` makes sure that only the time series based features are
filtered. This unit tests checks that
"""
augmenter = RelevantFeatureAugmenter(kind_to_fc_parameters=self.kind_to_fc_parameters,
filter_only_tsfresh_features=True,
column_value="val", column_id="id", column_sort="sort", column_kind="kind")
y = pd.Series({10: 1, 500: 0})
X = pd.DataFrame(index=[10, 500])
X["pre_feature"] = 0
augmenter.set_timeseries_container(self.test_df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(sum(["pre_feature" == column for column in transformed_X.columns]), 1)
def test_evaluate_only_added_features_false(self):
"""
The boolean flag `evaluate_only_extracted_features` makes sure that only the time series based features are
filtered. This unit tests checks that
"""
augmenter = RelevantFeatureAugmenter(kind_to_fc_parameters=self.kind_to_fc_parameters,
filter_only_tsfresh_features=False,
column_value="val", column_id="id", column_sort="sort", column_kind="kind")
df, y = self.create_test_data_sample_with_target()
X = pd.DataFrame(index=np.unique(df.id))
X["pre_drop"] = 0
X["pre_keep"] = y
augmenter.set_timeseries_container(df)
augmenter.fit(X, y)
transformed_X = augmenter.transform(X.copy())
self.assertEqual(sum(["pre_keep" == column for column in transformed_X.columns]), 1)
self.assertEqual(sum(["pre_drop" == column for column in transformed_X.columns]), 0)