def run(self, store) -> Report:
df1, df2 = store[self.metadata_precalculation]
part1, part2 = self.adjust_dataset_sizes(df1, df2)
categorical_check = CategoricalStatisticalCheck()
numerical_check = NumericalStatisticalCheck()
pvalues = pd.DataFrame(columns=df1.columns, index=['pvalue'])
for column in df1.columns.levels[0]:
for mdtype in self.metadata_precalculation.text_metadata_types:
if mdtype.metadata_return_type() == ColumnType.categorical:
p = categorical_check.statistical_test(
part1[(column, mdtype.metadata_name())],
part2[(column, mdtype.metadata_name())])
elif mdtype.metadata_return_type() == ColumnType.numerical:
p = numerical_check.statistical_test(
part1[(column, mdtype.metadata_name())],
part2[(column, mdtype.metadata_name())])
else:
raise UnknownMetadataReturnColumnTypeError(mdtype)
pvalues[(column, mdtype.metadata_name())] = [p]
significant_columns = self.significant_columns(pvalues)
return StatisticalReport(
"Text Metadata Check",
examined_columns=sorted(df1.columns.levels[0]),
shifted_columns=sorted(significant_columns),
explanation=self.explain(pvalues),
explanation_header=self.explanation_header(
numerical_check.statistical_test_name(),
categorical_check.statistical_test_name(),
any_significant=len(significant_columns) > 0),
information={'test_results': pvalues},
figures=self.metadata_figure(pvalues, part1, part2))
def metadata_plot(figure, tile, column, mdtype, df1, df2):
col_mdtype_tuple = (column, mdtype.metadata_name())
if mdtype.metadata_return_type() == ColumnType.categorical:
CategoricalStatisticalCheck.column_plot(figure, tile,
col_mdtype_tuple, df1, df2)
elif mdtype.metadata_return_type() == ColumnType.numerical:
NumericalStatisticalCheck.column_plot(figure, tile,
col_mdtype_tuple, df1, df2)
else:
raise UnknownMetadataReturnColumnTypeError(mdtype)
def test_size_adjustment(self):
df1 = pd.DataFrame([0] * 20)
df2 = pd.DataFrame([0] * 12)
for check, solution in [(CategoricalStatisticalCheck(sample_size=10), (10, 10)),
(CategoricalStatisticalCheck(sample_size=15), (15, 12)),
(NumericalStatisticalCheck(use_equal_dataset_sizes=True), (12, 12)),
(TextMetadataStatisticalCheck(sample_size=15, use_equal_dataset_sizes=True), (12, 12))]:
with self.subTest(solution=solution):
result1, result2 = check.adjust_dataset_sizes(df1, df2)
self.assertEqual(len(result1), solution[0])
self.assertEqual(len(result2), solution[1])
def test_paired_total_ratios_figure_looks_right(self, mock_plt):
mock_figure = MagicMock(autospec=Figure)
mock_axes = MagicMock(autospec=Axes)
with mock.patch.object(pd.DataFrame, 'plot') as mock_plot:
CategoricalStatisticalCheck.paired_total_ratios_plot(
mock_figure, mock_axes, 'vaccination_reaction',
self.df1_significant, self.df2_significant)
self.assertTrue(mock_plot.called)
self.assertTrue(mock_plot.return_value.set_title.called)
self.assertTrue(mock_plot.return_value.set_xlabel.called)
self.assertTrue(mock_plot.return_value.set_ylabel.called)
self.assertTrue(mock_plot.return_value.invert_yaxis.called)
self.assertFalse(mock_plt.show.called)
def test_column_tuples_are_handled_by_categorical_visualization(self):
columns = ['text']
metadata_names = ['category']
cols = pd.MultiIndex.from_product([columns, metadata_names], names=['column', 'metadata'])
df1 = pd.DataFrame(columns=cols)
df2 = pd.DataFrame(columns=cols)
df1[('text', 'category')] = ['latin' * 3]
df2[('text', 'category')] = ['arabic' * 3]
mock_figure = MagicMock(autospec=Figure)
mock_axes = MagicMock(autospec=Axes)
with mock.patch.object(categorical_statistical_check.vis, 'plot_categorical_horizontal_ratio_histogram',
return_value=mock_axes):
CategoricalStatisticalCheck.paired_total_ratios_plot(mock_figure, mock_axes, ('text', 'category'), df1, df2)
mock_axes.set_title.assert_called_once_with("Column: '('text', 'category')'", fontsize='x-large')
def test_figure_function_is_collected(self):
df1 = pd.DataFrame.from_dict({
'col1': ['value'] * 100,
'col2': ['value'] * 100
})
df2 = pd.DataFrame.from_dict({
'col1': ['value'] * 200,
'col2': ['value'] * 200
})
check = CategoricalStatisticalCheck()
for solution, sig_cols in [(1, ['col1', 'col2']), (1, ['col1']),
(0, [])]:
with self.subTest(solution=solution, sig_cols=sig_cols):
result = check.column_figure(significant_columns=sig_cols,
df1=df1,
df2=df2)
self.assertEqual(solution, len(result))
def test_all_plot_functions_are_called_and_plot_is_shown(self, mock_plt):
plot_data = [PlotData(MagicMock(), 1), PlotData(MagicMock(), 2), PlotData(MagicMock(), 3)]
for check, height in [(CategoricalStatisticalCheck(), 30.0), (NumericalStatisticalCheck(), 30.0)]:
with self.subTest(check=check):
check.plot_all_columns(plot_data)
mock_plt.figure.assert_called_with(figsize=(12, height), tight_layout=True)
for func, rows in plot_data:
self.assertTrue(func.called)
mock_plt.show.assert_called_with()
def test_plot_type(self, mock_plot, mock_subplot):
figure = MagicMock(autospec=Figure)
tile = MagicMock()
CategoricalStatisticalCheck().column_plot(figure, tile,
'vaccination_reaction',
self.df1_significant,
self.df2_significant)
mock_subplot.assert_called_with(figure, tile)
self.assertTrue(mock_plot.called)
def test_column_order_in_report(self):
df1 = pd.DataFrame([[1, 0]] * 10, columns=['abc', 'def'])
df2 = pd.DataFrame([[0, 1]] * 10, columns=['abc', 'def'])
store = Store(df1, df2)
for check in [CategoricalStatisticalCheck(), NumericalStatisticalCheck()]:
with self.subTest(check=check):
result = check.run(store)
self.assertEqual('abc', result.examined_columns[0])
self.assertEqual('abc', result.shifted_columns[0])
self.assertEqual(result.examined_columns, result.shifted_columns)
def test_compliance_with_detector(self):
df1 = pd.DataFrame([0] * 10)
df2 = pd.DataFrame([0] * 10)
detector = Detector(df1=df1, df2=df2, log_print=False)
detector.run(CategoricalStatisticalCheck())
self.assertEqual(1, len(detector.check_reports[0].examined_columns))
self.assertEqual(0, len(detector.check_reports[0].shifted_columns))
self.assertEqual(0, len(detector.check_reports[0].explanation))
assert_frame_equal(
pd.DataFrame([1.0], index=['pvalue']),
detector.check_reports[0].information['test_results'])
def test_correct_number_of_plot_functions(self):
for sig_cols in [['col1', 'col2'], ['col1'], []]:
with self.subTest(sig_cols=sig_cols):
result = CategoricalStatisticalCheck().plot_data(sig_cols, MagicMock(), MagicMock())
self.assertEqual(len(sig_cols), len(result))
def test_grid_is_created(self, mock_grid, mock_plt_figure):
plot_data = [PlotData(MagicMock(), 1), PlotData(MagicMock(), 2), PlotData(MagicMock(), 3)]
for check, shape in [(CategoricalStatisticalCheck(), (6, 1)), (NumericalStatisticalCheck(), (6, 1))]:
with self.subTest(check=check):
check.plot_all_columns(plot_data)
mock_grid.GridSpec.assert_called_with(shape[0], shape[1])
def test_significant(self):
store = Store(self.df1_significant, self.df2_significant)
result = CategoricalStatisticalCheck().run(store)
self.assertEqual(1, len(result.examined_columns))
self.assertEqual(1, len(result.shifted_columns))
self.assertEqual(1, len(result.explanation))