def test_incompatible_target(self) -> None:
dataframe = self._load_data(bad_features=[4])
hyperparams_class = MIRanking.metadata.query()["primitive_code"][
"class_type_arguments"
]["Hyperparams"]
hyperparams = hyperparams_class.defaults().replace({"target_col_index": 4})
mi_ranking = MIRanking(hyperparams=hyperparams)
result_dataframe = mi_ranking.produce(inputs=dataframe).value
# verify the output
self.assertListEqual(list(result_dataframe["idx"]), [])
self.assertListEqual(list(result_dataframe["name"]), [])
self.assertListEqual(list(result_dataframe["rank"]), [])
def test_continuous_target(self) -> None:
dataframe = self._load_data()
hyperparams_class = MIRanking.metadata.query()["primitive_code"][
"class_type_arguments"
]["Hyperparams"]
hyperparams = hyperparams_class.defaults().replace({"target_col_index": 2})
mi_ranking = MIRanking(hyperparams=hyperparams)
result_dataframe = mi_ranking.produce(inputs=dataframe).value
# verify the output
self.assertListEqual(list(result_dataframe["idx"]), [1, 5, 4, 3])
self.assertListEqual(
list(result_dataframe["name"]), ["alpha", "echo", "delta", "charlie"]
)
expected_ranks = [1.0, 0.930536, 7.316753e-16, 0.0]
for i, r in enumerate(result_dataframe["rank"]):
self.assertAlmostEqual(r, expected_ranks[i], places=6)
def test_unique_categorical_removed(self) -> None:
dataframe = self._load_data()
hyperparams_class = MIRanking.metadata.query()["primitive_code"][
"class_type_arguments"
]["Hyperparams"]
hyperparams = hyperparams_class.defaults().replace({"target_col_index": 1})
dataframe.insert(len(dataframe.columns), "removed", [1, 2, 3, 4, 5, 6, 7, 8, 9])
dataframe.metadata = dataframe.metadata.add_semantic_type(
(metadata_base.ALL_ELEMENTS, 1),
"https://metadata.datadrivendiscovery.org/types/CategoricalData",
)
mi_ranking = MIRanking(hyperparams=hyperparams)
result_dataframe = mi_ranking.produce(inputs=dataframe).value
# verify the output
self.assertListEqual(list(result_dataframe["idx"]), [2, 5, 4, 3])
self.assertListEqual(
list(result_dataframe["name"]), ["bravo", "echo", "delta", "charlie"]
)
def test_discrete_target_rank_in_metadata(self) -> None:
dataframe = self._load_data()
hyperparams_class = MIRanking.metadata.query()["primitive_code"][
"class_type_arguments"
]["Hyperparams"]
hyperparams = hyperparams_class.defaults().replace(
{"target_col_index": 1, "return_as_metadata": True}
)
mi_ranking = MIRanking(hyperparams=hyperparams)
result_dataframe = mi_ranking.produce(inputs=dataframe).value
expected_ranks = [1.0, 1.0, 1.0, 0.0]
ranked_cols = [2, 5, 4, 3]
for i in range(len(ranked_cols)):
self.assertAlmostEqual(
result_dataframe.metadata.query(
(metadata_base.ALL_ELEMENTS, ranked_cols[i])
).get("rank"),
expected_ranks[i],
places=6,
)
def test_full_mutual_info(self) -> None:
dataframe = self._load_data(alpha_class=float, charlie_class=float)
hyperparams_class = MIRanking.metadata.query()["primitive_code"][
"class_type_arguments"
]["Hyperparams"]
hyperparams = hyperparams_class.defaults().replace({"target_col_index": 2})
mi_ranking = MIRanking(hyperparams=hyperparams)
# dataframe.drop(columns=['charlie', 'delta', 'echo'], inplace=True)
# dataframe['bravo'][dataframe['bravo'] == 4] = 3
dataframe["bravo"] = [1, 0.5, 10, 6.7, 6.9, 2.3, 5.5, 7.3, 9]
dataframe["alpha"] = [1, 0.5, 10, 6.7, 6.9, 2.3, 5.5, 7.3, 9]
dataframe["charlie"] = [2, 2.5, 9, 5.7, 7.9, 1.3, 6.5, 8.3, 8]
result_dataframe = mi_ranking.produce(inputs=dataframe).value
# verify the output
self.assertListEqual(list(result_dataframe["idx"]), [1, 3, 4, 5])
self.assertListEqual(
list(result_dataframe["name"]), ["alpha", "charlie", "delta", "echo"]
)
expected_ranks = [1.0, 0.665361, 0.0, 0.0]
for i, r in enumerate(result_dataframe["rank"]):
self.assertAlmostEqual(r, expected_ranks[i], places=6)