class ClassifierBaseExplainerTests(unittest.TestCase):
def setUp(self):
X_train, y_train, X_test, y_test = titanic_survive()
train_names, test_names = titanic_names()
model = RandomForestClassifier(n_estimators=5, max_depth=2)
model.fit(X_train, y_train)
self.explainer = ClassifierExplainer(
model,
X_test,
y_test,
roc_auc_score,
cats=[{
'Gender': ['Sex_female', 'Sex_male', 'Sex_nan']
}, 'Deck', 'Embarked'],
target='Survival',
labels=['Not survived', 'Survived'],
idxs=test_names)
def test_explainer_len(self):
self.assertEqual(len(self.explainer), len(titanic_survive()[2]))
def test_int_idx(self):
self.assertEqual(self.explainer.get_int_idx(titanic_names()[1][0]), 0)
def test_random_index(self):
self.assertIsInstance(self.explainer.random_index(), int)
self.assertIsInstance(self.explainer.random_index(return_str=True),
str)
def test_preds(self):
self.assertIsInstance(self.explainer.preds, np.ndarray)
def test_pred_percentiles(self):
self.assertIsInstance(self.explainer.pred_percentiles, np.ndarray)
def test_columns_ranked_by_shap(self):
self.assertIsInstance(self.explainer.columns_ranked_by_shap(), list)
self.assertIsInstance(self.explainer.columns_ranked_by_shap(cats=True),
list)
def test_equivalent_col(self):
self.assertEqual(self.explainer.equivalent_col("Sex_female"), "Gender")
self.assertEqual(self.explainer.equivalent_col("Gender"), "Sex_female")
self.assertEqual(self.explainer.equivalent_col("Deck"), "Deck_A")
self.assertEqual(self.explainer.equivalent_col("Deck_A"), "Deck")
self.assertIsNone(self.explainer.equivalent_col("random"))
def test_get_col(self):
self.assertIsInstance(self.explainer.get_col("Gender"), pd.Series)
self.assertEqual(self.explainer.get_col("Gender").dtype, "object")
self.assertIsInstance(self.explainer.get_col("Deck"), pd.Series)
self.assertEqual(self.explainer.get_col("Deck").dtype, "object")
self.assertIsInstance(self.explainer.get_col("Age"), pd.Series)
self.assertEqual(self.explainer.get_col("Age").dtype, np.float)
def test_permutation_importances(self):
self.assertIsInstance(self.explainer.permutation_importances,
pd.DataFrame)
self.assertIsInstance(self.explainer.permutation_importances_cats,
pd.DataFrame)
def test_X_cats(self):
self.assertIsInstance(self.explainer.X_cats, pd.DataFrame)
def test_columns_cats(self):
self.assertIsInstance(self.explainer.columns_cats, list)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics_markdown(), str)
def test_mean_abs_shap_df(self):
self.assertIsInstance(self.explainer.mean_abs_shap_df(), pd.DataFrame)
def test_top_interactions(self):
self.assertIsInstance(self.explainer.shap_top_interactions("Age"),
list)
self.assertIsInstance(
self.explainer.shap_top_interactions("Age", topx=4), list)
self.assertIsInstance(
self.explainer.shap_top_interactions("Age", cats=True), list)
self.assertIsInstance(
self.explainer.shap_top_interactions("Gender", cats=True), list)
def test_permutation_importances_df(self):
self.assertIsInstance(self.explainer.permutation_importances_df(),
pd.DataFrame)
self.assertIsInstance(
self.explainer.permutation_importances_df(topx=3), pd.DataFrame)
self.assertIsInstance(
self.explainer.permutation_importances_df(cats=True), pd.DataFrame)
self.assertIsInstance(
self.explainer.permutation_importances_df(cutoff=0.01),
pd.DataFrame)
def test_contrib_df(self):
self.assertIsInstance(self.explainer.contrib_df(0), pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, cats=False),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, topx=3),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, sort='high-to-low'),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, sort='low-to-high'),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, sort='importance'),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_df(X_row=self.explainer.X.iloc[[0]]),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_df(X_row=self.explainer.X_cats.iloc[[0]]),
pd.DataFrame)
def test_contrib_summary_df(self):
self.assertIsInstance(self.explainer.contrib_summary_df(0),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_summary_df(0, cats=False),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_summary_df(0, topx=3),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_summary_df(0, round=3),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_summary_df(0, sort='low-to-high'),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_summary_df(0, sort='high-to-low'),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_summary_df(0, sort='importance'),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_summary_df(
X_row=self.explainer.X.iloc[[0]]), pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_summary_df(
X_row=self.explainer.X_cats.iloc[[0]]), pd.DataFrame)
def test_shap_base_value(self):
self.assertIsInstance(self.explainer.shap_base_value,
(np.floating, float))
def test_shap_values_shape(self):
self.assertTrue(
self.explainer.shap_values.shape == (len(self.explainer),
len(self.explainer.columns)))
def test_shap_values(self):
self.assertIsInstance(self.explainer.shap_values, np.ndarray)
self.assertIsInstance(self.explainer.shap_values_cats, np.ndarray)
def test_shap_interaction_values(self):
self.assertIsInstance(self.explainer.shap_interaction_values,
np.ndarray)
self.assertIsInstance(self.explainer.shap_interaction_values_cats,
np.ndarray)
def test_mean_abs_shap(self):
self.assertIsInstance(self.explainer.mean_abs_shap, pd.DataFrame)
self.assertIsInstance(self.explainer.mean_abs_shap_cats, pd.DataFrame)
def test_calculate_properties(self):
self.explainer.calculate_properties()
def test_shap_interaction_values_by_col(self):
self.assertIsInstance(
self.explainer.shap_interaction_values_by_col("Age"), np.ndarray)
self.assertEquals(
self.explainer.shap_interaction_values_by_col("Age").shape,
self.explainer.shap_values.shape)
self.assertEquals(
self.explainer.shap_interaction_values_by_col("Age",
cats=True).shape,
self.explainer.shap_values_cats.shape)
def test_pdp_result(self):
self.assertIsInstance(self.explainer.get_pdp_result("Age"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Gender"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Age", index=0),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Gender", index=0),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(
self.explainer.get_pdp_result("Age",
X_row=self.explainer.X.iloc[[0]]),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(
self.explainer.get_pdp_result(
"Age", X_row=self.explainer.X_cats.iloc[[0]]),
pdpbox.pdp.PDPIsolate)
def test_get_dfs(self):
cols_df, shap_df, contribs_df = self.explainer.get_dfs()
self.assertIsInstance(cols_df, pd.DataFrame)
self.assertIsInstance(shap_df, pd.DataFrame)
self.assertIsInstance(contribs_df, pd.DataFrame)
def test_plot_importances(self):
fig = self.explainer.plot_importances()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_importances(kind='permutation')
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_importances(topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_importances(cats=True)
self.assertIsInstance(fig, go.Figure)
def test_plot_interactions(self):
fig = self.explainer.plot_interactions("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_interactions("Sex_female")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_interactions("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_interactions("Gender")
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_contributions(self):
fig = self.explainer.plot_shap_contributions(0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, cats=False)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, cutoff=0.05)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, sort='high-to-low')
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, sort='low-to-high')
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, sort='importance')
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(
X_row=self.explainer.X.iloc[[0]], sort='importance')
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(
X_row=self.explainer.X_cats.iloc[[0]], sort='importance')
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_summary(self):
fig = self.explainer.plot_shap_summary()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_summary(topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_summary(cats=True)
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_interaction_summary(self):
fig = self.explainer.plot_shap_interaction_summary("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Age", topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Sex_female",
topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Gender")
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_dependence(self):
fig = self.explainer.plot_shap_dependence("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Sex_female")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age", "Gender")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Sex_female", "Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age", highlight_index=0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Gender", highlight_index=0)
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_interaction(self):
fig = self.explainer.plot_shap_dependence("Age", "Sex_female")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Sex_female", "Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Gender", "Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age", "Gender")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age",
"Sex_female",
highlight_index=0)
self.assertIsInstance(fig, go.Figure)
def test_plot_pdp(self):
fig = self.explainer.plot_pdp("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Gender")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Gender", index=0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Age", index=0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Age", X_row=self.explainer.X.iloc[[0]])
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Age",
X_row=self.explainer.X_cats.iloc[[0]])
self.assertIsInstance(fig, go.Figure)
def test_yaml(self):
yaml = self.explainer.to_yaml()
self.assertIsInstance(yaml, str)
class LogisticRegressionTests(unittest.TestCase):
def setUp(self):
X_train, y_train, X_test, y_test = titanic_survive()
train_names, test_names = titanic_names()
model = LogisticRegression()
model.fit(X_train, y_train)
self.explainer = ClassifierExplainer(
model,
X_test,
y_test,
roc_auc_score,
shap='linear',
cats=['Sex', 'Cabin', 'Embarked'],
labels=['Not survived', 'Survived'],
idxs=test_names)
def test_preds(self):
self.assertIsInstance(self.explainer.preds, np.ndarray)
def test_pred_percentiles(self):
self.assertIsInstance(self.explainer.pred_percentiles, np.ndarray)
def test_columns_ranked_by_shap(self):
self.assertIsInstance(self.explainer.columns_ranked_by_shap(), list)
self.assertIsInstance(self.explainer.columns_ranked_by_shap(cats=True),
list)
def test_permutation_importances(self):
self.assertIsInstance(self.explainer.permutation_importances,
pd.DataFrame)
self.assertIsInstance(self.explainer.permutation_importances_cats,
pd.DataFrame)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics_markdown(), str)
def test_mean_abs_shap_df(self):
self.assertIsInstance(self.explainer.mean_abs_shap_df(), pd.DataFrame)
def test_contrib_df(self):
self.assertIsInstance(self.explainer.contrib_df(0), pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, cats=False),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, topx=3),
pd.DataFrame)
def test_shap_base_value(self):
self.assertIsInstance(self.explainer.shap_base_value,
(np.floating, float))
def test_shap_values_shape(self):
self.assertTrue(
self.explainer.shap_values.shape == (len(self.explainer),
len(self.explainer.columns)))
def test_shap_values(self):
self.assertIsInstance(self.explainer.shap_values, np.ndarray)
self.assertIsInstance(self.explainer.shap_values_cats, np.ndarray)
def test_mean_abs_shap(self):
self.assertIsInstance(self.explainer.mean_abs_shap, pd.DataFrame)
self.assertIsInstance(self.explainer.mean_abs_shap_cats, pd.DataFrame)
def test_calculate_properties(self):
self.explainer.calculate_properties(include_interactions=False)
def test_pdp_result(self):
self.assertIsInstance(self.explainer.get_pdp_result("Age"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Sex"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Age", index=0),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Sex", index=0),
pdpbox.pdp.PDPIsolate)
def test_pos_label(self):
self.explainer.pos_label = 1
self.explainer.pos_label = "Not survived"
self.assertIsInstance(self.explainer.pos_label, int)
self.assertIsInstance(self.explainer.pos_label_str, str)
self.assertEquals(self.explainer.pos_label, 0)
self.assertEquals(self.explainer.pos_label_str, "Not survived")
def test_get_prop_for_label(self):
self.explainer.pos_label = 1
tmp = self.explainer.pred_percentiles
self.explainer.pos_label = 0
self.assertTrue(
np.alltrue(
self.explainer.get_prop_for_label("pred_percentiles", 1) ==
tmp))
def test_pred_probas(self):
self.assertIsInstance(self.explainer.pred_probas, np.ndarray)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics(cutoff=0.9), dict)
def test_precision_df(self):
self.assertIsInstance(self.explainer.precision_df(), pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(multiclass=True),
pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(quantiles=4),
pd.DataFrame)
def test_lift_curve_df(self):
self.assertIsInstance(self.explainer.lift_curve_df(), pd.DataFrame)
def test_prediction_result_markdown(self):
self.assertIsInstance(self.explainer.prediction_result_markdown(0),
str)
class CatBoostClassifierTests(unittest.TestCase):
def setUp(self):
X_train, y_train, X_test, y_test = titanic_survive()
train_names, test_names = titanic_names()
model = CatBoostClassifier(iterations=100, learning_rate=0.1, verbose=0)
model.fit(X_train, y_train)
self.explainer = ClassifierExplainer(
model, X_test, y_test, roc_auc_score,
shap='tree',
cats=['Sex', 'Cabin', 'Embarked'],
labels=['Not survived', 'Survived'],
idxs=test_names)
def test_preds(self):
self.assertIsInstance(self.explainer.preds, np.ndarray)
def test_pred_probas(self):
self.assertIsInstance(self.explainer.pred_probas, np.ndarray)
def test_permutation_importances(self):
self.assertIsInstance(self.explainer.permutation_importances, pd.DataFrame)
self.assertIsInstance(self.explainer.permutation_importances_cats, pd.DataFrame)
def test_shap_base_value(self):
self.assertIsInstance(self.explainer.shap_base_value, (np.floating, float))
def test_shap_values_all_probabilities(self):
self.assertTrue(self.explainer.shap_base_value >= 0)
self.assertTrue(self.explainer.shap_base_value <= 1)
self.assertTrue(np.all(self.explainer.shap_values.sum(axis=1) + self.explainer.shap_base_value >= 0))
self.assertTrue(np.all(self.explainer.shap_values.sum(axis=1) + self.explainer.shap_base_value <= 1))
def test_shap_values_shape(self):
self.assertTrue(self.explainer.shap_values.shape == (len(self.explainer), len(self.explainer.columns)))
def test_shap_values(self):
self.assertIsInstance(self.explainer.shap_values, np.ndarray)
self.assertIsInstance(self.explainer.shap_values_cats, np.ndarray)
# @unittest.expectedFailure
# def test_shap_interaction_values(self):
# self.assertIsInstance(self.explainer.shap_interaction_values, np.ndarray)
# self.assertIsInstance(self.explainer.shap_interaction_values_cats, np.ndarray)
def test_mean_abs_shap(self):
self.assertIsInstance(self.explainer.mean_abs_shap, pd.DataFrame)
self.assertIsInstance(self.explainer.mean_abs_shap_cats, pd.DataFrame)
def test_calculate_properties(self):
self.explainer.calculate_properties(include_interactions=False)
def test_pdp_result(self):
self.assertIsInstance(self.explainer.get_pdp_result("Age"), pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Sex"), pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Age", index=0), pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Sex", index=0), pdpbox.pdp.PDPIsolate)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics(cutoff=0.9), dict)
def test_precision_df(self):
self.assertIsInstance(self.explainer.precision_df(), pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(multiclass=True), pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(quantiles=4), pd.DataFrame)
def test_lift_curve_df(self):
self.assertIsInstance(self.explainer.lift_curve_df(), pd.DataFrame)
def test_prediction_result_markdown(self):
self.assertIsInstance(self.explainer.prediction_result_markdown(0), str)
class ClassifierBaseExplainerTestsPipeline(unittest.TestCase):
def setUp(self):
#X, y = fetch_openml("titanic", version=1, as_frame=True, return_X_y=True)
df = pd.read_csv(Path.cwd() / "tests" / "cli_assets" /
"pipeline_data.csv")
X = df[['age', 'fare', 'embarked', 'sex', 'pclass']]
y = df['survived'].astype(int)
numeric_features = ['age', 'fare']
numeric_transformer = Pipeline(
steps=[('imputer', SimpleImputer(
strategy='median')), ('scaler', StandardScaler())])
categorical_features = ['embarked', 'sex', 'pclass']
categorical_transformer = Pipeline(
steps=[('imputer', SimpleImputer(
strategy='most_frequent')), ('ordinal', OrdinalEncoder())])
preprocessor = ColumnTransformer(
transformers=[('num', numeric_transformer, numeric_features),
('cat', categorical_transformer,
categorical_features)])
# Append classifier to preprocessing pipeline.
# Now we have a full prediction pipeline.
clf = Pipeline(
steps=[('preprocessor',
preprocessor), ('classifier', RandomForestClassifier())])
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.2)
clf.fit(X_train, y_train)
self.explainer = ClassifierExplainer(clf, X_test, y_test)
def test_columns_ranked_by_shap(self):
self.assertIsInstance(self.explainer.columns_ranked_by_shap(), list)
def test_permutation_importances(self):
self.assertIsInstance(self.explainer.permutation_importances,
pd.DataFrame)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics_descriptions(), dict)
def test_mean_abs_shap_df(self):
self.assertIsInstance(self.explainer.mean_abs_shap_df(), pd.DataFrame)
def test_contrib_df(self):
self.assertIsInstance(self.explainer.contrib_df(0), pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, cats=False),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, topx=3),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, sort='high-to-low'),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, sort='low-to-high'),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, sort='importance'),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_df(X_row=self.explainer.X.iloc[[0]]),
pd.DataFrame)
self.assertIsInstance(
self.explainer.contrib_df(X_row=self.explainer.X_cats.iloc[[0]]),
pd.DataFrame)
def test_shap_base_value(self):
self.assertIsInstance(self.explainer.shap_base_value,
(np.floating, float))
def test_shap_values_shape(self):
self.assertTrue(
self.explainer.shap_values.shape == (len(self.explainer),
len(self.explainer.columns)))
def test_shap_values(self):
self.assertIsInstance(self.explainer.shap_values, np.ndarray)
self.assertIsInstance(self.explainer.shap_values_cats, np.ndarray)
def test_pdp_result(self):
self.assertIsInstance(self.explainer.get_pdp_result("age"),
pdpbox.pdp.PDPIsolate)
class XGBCLassifierTests(unittest.TestCase):
def setUp(self):
X_train, y_train, X_test, y_test = titanic_survive()
train_names, test_names = titanic_names()
model = XGBClassifier()
model.fit(X_train, y_train)
self.explainer = ClassifierExplainer(
model,
X_test,
y_test,
cats=[{
'Gender': ['Sex_female', 'Sex_male', 'Sex_nan']
}, 'Deck', 'Embarked'],
labels=['Not survived', 'Survived'])
def test_preds(self):
self.assertIsInstance(self.explainer.preds, np.ndarray)
def test_pred_probas(self):
self.assertIsInstance(self.explainer.pred_probas, np.ndarray)
def test_permutation_importances(self):
self.assertIsInstance(self.explainer.permutation_importances,
pd.DataFrame)
self.assertIsInstance(self.explainer.permutation_importances_cats,
pd.DataFrame)
def test_shap_base_value(self):
self.assertIsInstance(self.explainer.shap_base_value,
(np.floating, float))
def test_shap_values_shape(self):
self.assertTrue(
self.explainer.shap_values.shape == (len(self.explainer),
len(self.explainer.columns)))
def test_shap_values(self):
self.assertIsInstance(self.explainer.shap_values, np.ndarray)
self.assertIsInstance(self.explainer.shap_values_cats, np.ndarray)
def test_shap_values_all_probabilities(self):
self.assertTrue(self.explainer.shap_base_value >= 0)
self.assertTrue(self.explainer.shap_base_value <= 1)
self.assertTrue(
np.all(
self.explainer.shap_values.sum(axis=1) +
self.explainer.shap_base_value >= 0))
self.assertTrue(
np.all(
self.explainer.shap_values.sum(axis=1) +
self.explainer.shap_base_value <= 1))
# def test_shap_interaction_values(self):
# self.assertIsInstance(self.explainer.shap_interaction_values, np.ndarray)
# self.assertIsInstance(self.explainer.shap_interaction_values_cats, np.ndarray)
def test_mean_abs_shap(self):
self.assertIsInstance(self.explainer.mean_abs_shap, pd.DataFrame)
self.assertIsInstance(self.explainer.mean_abs_shap_cats, pd.DataFrame)
def test_calculate_properties(self):
self.explainer.calculate_properties(include_interactions=False)
def test_pdp_result(self):
self.assertIsInstance(self.explainer.get_pdp_result("Age"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Gender"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Age", index=0),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Gender", index=0),
pdpbox.pdp.PDPIsolate)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics(cutoff=0.9), dict)
def test_precision_df(self):
self.assertIsInstance(self.explainer.precision_df(), pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(multiclass=True),
pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(quantiles=4),
pd.DataFrame)
def test_lift_curve_df(self):
self.assertIsInstance(self.explainer.lift_curve_df(), pd.DataFrame)
def test_prediction_result_markdown(self):
self.assertIsInstance(self.explainer.prediction_result_markdown(0),
str)
class MultiClassClassifierBunchTests(unittest.TestCase):
def setUp(self):
X_train, y_train, X_test, y_test = titanic_embarked()
train_names, test_names = titanic_names()
model = RandomForestClassifier(n_estimators=5, max_depth=2)
model.fit(X_train, y_train)
self.explainer = ClassifierExplainer(
model,
X_test,
y_test,
cats=[{
'Gender': ['Sex_female', 'Sex_male', 'Sex_nan']
}, 'Deck'],
idxs=test_names,
labels=['Queenstown', 'Southampton', 'Cherbourg'])
def test_preds(self):
self.assertIsInstance(self.explainer.preds, np.ndarray)
def test_pred_percentiles(self):
self.assertIsInstance(self.explainer.pred_percentiles, np.ndarray)
def test_columns_ranked_by_shap(self):
self.assertIsInstance(self.explainer.columns_ranked_by_shap(), list)
self.assertIsInstance(self.explainer.columns_ranked_by_shap(cats=True),
list)
def test_equivalent_col(self):
self.assertEqual(self.explainer.equivalent_col("Sex_female"), "Gender")
self.assertEqual(self.explainer.equivalent_col("Gender"), "Sex_female")
self.assertIsNone(self.explainer.equivalent_col("random"))
def test_get_col(self):
self.assertIsInstance(self.explainer.get_col("Gender"), pd.Series)
self.assertEqual(self.explainer.get_col("Gender").dtype, "object")
self.assertIsInstance(self.explainer.get_col("Age"), pd.Series)
self.assertEqual(self.explainer.get_col("Age").dtype, np.float)
def test_permutation_importances(self):
self.assertIsInstance(self.explainer.permutation_importances,
pd.DataFrame)
self.assertIsInstance(self.explainer.permutation_importances_cats,
pd.DataFrame)
def test_X_cats(self):
self.assertIsInstance(self.explainer.X_cats, pd.DataFrame)
def test_columns_cats(self):
self.assertIsInstance(self.explainer.columns_cats, list)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics_markdown(), str)
def test_mean_abs_shap_df(self):
self.assertIsInstance(self.explainer.mean_abs_shap_df(), pd.DataFrame)
def test_top_interactions(self):
self.assertIsInstance(self.explainer.shap_top_interactions("Age"),
list)
self.assertIsInstance(
self.explainer.shap_top_interactions("Age", topx=4), list)
self.assertIsInstance(
self.explainer.shap_top_interactions("Age", cats=True), list)
self.assertIsInstance(
self.explainer.shap_top_interactions("Gender", cats=True), list)
def test_permutation_importances_df(self):
self.assertIsInstance(self.explainer.permutation_importances_df(),
pd.DataFrame)
self.assertIsInstance(
self.explainer.permutation_importances_df(topx=3), pd.DataFrame)
self.assertIsInstance(
self.explainer.permutation_importances_df(cats=True), pd.DataFrame)
self.assertIsInstance(
self.explainer.permutation_importances_df(cutoff=0.01),
pd.DataFrame)
def test_contrib_df(self):
self.assertIsInstance(self.explainer.contrib_df(0), pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, cats=False),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_df(0, topx=3),
pd.DataFrame)
def test_contrib_summary_df(self):
self.assertIsInstance(self.explainer.contrib_summary_df(0),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_summary_df(0, cats=False),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_summary_df(0, topx=3),
pd.DataFrame)
self.assertIsInstance(self.explainer.contrib_summary_df(0, round=3),
pd.DataFrame)
def test_shap_base_value(self):
self.assertIsInstance(self.explainer.shap_base_value,
(np.floating, float))
def test_shap_values_shape(self):
self.assertTrue(
self.explainer.shap_values.shape == (len(self.explainer),
len(self.explainer.columns)))
def test_shap_values(self):
self.assertIsInstance(self.explainer.shap_values, np.ndarray)
self.assertIsInstance(self.explainer.shap_values_cats, np.ndarray)
def test_shap_interaction_values(self):
self.assertIsInstance(self.explainer.shap_interaction_values,
np.ndarray)
self.assertIsInstance(self.explainer.shap_interaction_values_cats,
np.ndarray)
def test_mean_abs_shap(self):
self.assertIsInstance(self.explainer.mean_abs_shap, pd.DataFrame)
self.assertIsInstance(self.explainer.mean_abs_shap_cats, pd.DataFrame)
def test_calculate_properties(self):
self.explainer.calculate_properties()
def test_shap_interaction_values_by_col(self):
self.assertIsInstance(
self.explainer.shap_interaction_values_by_col("Age"), np.ndarray)
self.assertEquals(
self.explainer.shap_interaction_values_by_col("Age").shape,
self.explainer.shap_values.shape)
self.assertEquals(
self.explainer.shap_interaction_values_by_col("Age",
cats=True).shape,
self.explainer.shap_values_cats.shape)
def test_pdp_result(self):
self.assertIsInstance(self.explainer.get_pdp_result("Age"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Gender"),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Age", index=0),
pdpbox.pdp.PDPIsolate)
self.assertIsInstance(self.explainer.get_pdp_result("Gender", index=0),
pdpbox.pdp.PDPIsolate)
def test_get_dfs(self):
cols_df, shap_df, contribs_df = self.explainer.get_dfs()
self.assertIsInstance(cols_df, pd.DataFrame)
self.assertIsInstance(shap_df, pd.DataFrame)
self.assertIsInstance(contribs_df, pd.DataFrame)
def test_plot_importances(self):
fig = self.explainer.plot_importances()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_importances(kind='permutation')
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_importances(topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_importances(cats=True)
self.assertIsInstance(fig, go.Figure)
def test_plot_interactions(self):
fig = self.explainer.plot_interactions("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_interactions("Sex_female")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_interactions("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_interactions("Gender")
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_contributions(self):
fig = self.explainer.plot_shap_contributions(0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, cats=False)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_contributions(0, cutoff=0.05)
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_summary(self):
fig = self.explainer.plot_shap_summary()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_summary(topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_summary(cats=True)
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_interaction_summary(self):
fig = self.explainer.plot_shap_interaction_summary("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Age", topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Sex_female",
topx=3)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_interaction_summary("Gender")
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_dependence(self):
fig = self.explainer.plot_shap_dependence("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Sex_female")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age", "Gender")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Sex_female", "Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age", highlight_index=0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Gender", highlight_index=0)
self.assertIsInstance(fig, go.Figure)
def test_plot_shap_interaction(self):
fig = self.explainer.plot_shap_dependence("Age", "Sex_female")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Sex_female", "Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Gender", "Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age", "Gender")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_shap_dependence("Age",
"Sex_female",
highlight_index=0)
self.assertIsInstance(fig, go.Figure)
def test_plot_pdp(self):
fig = self.explainer.plot_pdp("Age")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Gender")
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Gender", index=0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pdp("Age", index=0)
self.assertIsInstance(fig, go.Figure)
def test_pos_label(self):
self.explainer.pos_label = 1
self.explainer.pos_label = "Southampton"
self.assertIsInstance(self.explainer.pos_label, int)
self.assertIsInstance(self.explainer.pos_label_str, str)
self.assertEquals(self.explainer.pos_label, 1)
self.assertEquals(self.explainer.pos_label_str, "Southampton")
def test_get_prop_for_label(self):
self.explainer.pos_label = 1
tmp = self.explainer.pred_percentiles
self.explainer.pos_label = 0
self.assertTrue(
np.alltrue(
self.explainer.get_prop_for_label("pred_percentiles", 1) ==
tmp))
def test_pred_probas(self):
self.assertIsInstance(self.explainer.pred_probas, np.ndarray)
def test_metrics(self):
self.assertIsInstance(self.explainer.metrics(), dict)
self.assertIsInstance(self.explainer.metrics(cutoff=0.9), dict)
def test_precision_df(self):
self.assertIsInstance(self.explainer.precision_df(), pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(multiclass=True),
pd.DataFrame)
self.assertIsInstance(self.explainer.precision_df(quantiles=4),
pd.DataFrame)
def test_lift_curve_df(self):
self.assertIsInstance(self.explainer.lift_curve_df(), pd.DataFrame)
def test_prediction_result_markdown(self):
self.assertIsInstance(self.explainer.prediction_result_markdown(0),
str)
def test_calculate_properties(self):
self.explainer.calculate_properties()
def test_plot_precision(self):
fig = self.explainer.plot_precision()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_precision(multiclass=True)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_precision(quantiles=10, cutoff=0.5)
self.assertIsInstance(fig, go.Figure)
def test_plot_cumulative_precision(self):
fig = self.explainer.plot_cumulative_precision()
self.assertIsInstance(fig, go.Figure)
def test_plot_confusion_matrix(self):
fig = self.explainer.plot_confusion_matrix(normalized=False,
binary=False)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_confusion_matrix(normalized=False,
binary=True)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_confusion_matrix(normalized=True,
binary=False)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_confusion_matrix(normalized=True,
binary=True)
self.assertIsInstance(fig, go.Figure)
def test_plot_lift_curve(self):
fig = self.explainer.plot_lift_curve()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_lift_curve(percentage=True)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_lift_curve(cutoff=0.5)
self.assertIsInstance(fig, go.Figure)
def test_plot_lift_curve(self):
fig = self.explainer.plot_lift_curve()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_lift_curve(percentage=True)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_lift_curve(cutoff=0.5)
self.assertIsInstance(fig, go.Figure)
def test_plot_classification(self):
fig = self.explainer.plot_classification()
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_classification(percentage=True)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_classification(cutoff=0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_classification(cutoff=1)
self.assertIsInstance(fig, go.Figure)
def test_plot_roc_auc(self):
fig = self.explainer.plot_roc_auc(0.5)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_roc_auc(0.0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_roc_auc(1.0)
self.assertIsInstance(fig, go.Figure)
def test_plot_pr_auc(self):
fig = self.explainer.plot_pr_auc(0.5)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pr_auc(0.0)
self.assertIsInstance(fig, go.Figure)
fig = self.explainer.plot_pr_auc(1.0)
self.assertIsInstance(fig, go.Figure)
