def test_causal_manager_global_cohort_effects(self, housing_data):
train_df, test_df, target_feature = housing_data
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, None)
manager.add(['AveRooms'])
manager.compute()
id = manager.get()[0].id
X_test = test_df.drop(target_feature, axis=1)
causal_data = manager.request_global_cohort_effects(id, X_test)
self.verify_common_causal_data_attributes(causal_data)
assert hasattr(causal_data, 'global_effects')
EFFECTS_ATTRIBUTES = [
'point',
'stderr',
'zstat',
'ci_lower',
'ci_upper',
'p_value',
]
for effect in EFFECTS_ATTRIBUTES:
assert effect in causal_data.global_effects[0]
incorrect_query_id = "incorrect_query_id"
X_test = test_df.drop(target_feature, axis=1)
with pytest.raises(ValueError,
match="Failed to find causal result with ID: "
"incorrect_query_id"):
manager.request_global_cohort_effects(incorrect_query_id, X_test)
def _initialize_managers(self):
"""Initializes the managers.
Initialized the causal, counterfactual, error analysis
and explainer managers.
"""
self._causal_manager = CausalManager(
self.train, self.test, self.target_column,
self.task_type, self.categorical_features)
self._counterfactual_manager = CounterfactualManager(
model=self.model, train=self.train, test=self.test,
target_column=self.target_column, task_type=self.task_type,
categorical_features=self.categorical_features)
self._error_analysis_manager = ErrorAnalysisManager(
self.model, self.test, self.target_column,
self._classes,
self.categorical_features)
self._explainer_manager = ExplainerManager(
self.model, self.train, self.test,
self.target_column,
self._classes,
categorical_features=self.categorical_features)
self._managers = [self._causal_manager,
self._counterfactual_manager,
self._error_analysis_manager,
self._explainer_manager]
def test_causal_train_test_categories(self, parks_data):
train_df, test_df, target_feature = parks_data
test_df = test_df.copy()
test_df.loc[len(test_df.index)] = ['indiana', 301, 'trees', 78]
test_df.loc[len(test_df.index)] = ['indiana', 222, 'trees', 81]
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, ['state', 'attraction'])
message = ("Causal analysis requires that every category of "
"categorical features present in the test data be "
"also present in the train data. "
"Categories missing from train data: "
"{'state': \\['indiana'\\]}")
with pytest.raises(UserConfigValidationException, match=message):
manager.add(['state'],
skip_cat_limit_checks=True,
upper_bound_on_cat_expansion=50)
def causal_result(parks_data):
train_df, test_df, target_feature = parks_data
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, ['state', 'attraction'])
manager.add(['attraction'],
skip_cat_limit_checks=True,
upper_bound_on_cat_expansion=50)
manager.compute()
return manager.get()[0]
def test_categorical_policy(self, housing_data_categorical):
train_df, test_df, target_feature = housing_data_categorical
categoricals = train_df.select_dtypes(include=[object]).columns
# Just use categoricals to force categorical policy tree
new_features = list(categoricals) + [target_feature]
train_df = train_df[new_features]
test_df = test_df[new_features]
# Sample data for easier debug
test_df = test_df[:20]
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, categoricals)
result = manager.add(['HouseAge_CAT', 'Population_CAT'],
random_state=42)
manager.compute()
result = manager.get()[0]
dashboard_data = result._get_dashboard_data()
policies = dashboard_data['policies']
assert len(policies) > 0
for policy in policies:
tree = policy['policy_tree']
assert not tree['leaf']
assert tree['feature'] in categoricals
assert tree['right_comparison'] == 'eq'
is_very_old_comparison = tree['comparison_value'] == 'very-old'
is_high_comparison = tree['comparison_value'] == 'high'
assert is_very_old_comparison or is_high_comparison
def test_causal_no_categoricals(self, housing_data):
train_df, test_df, target_feature = housing_data
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, None)
manager.add(['AveRooms'])
manager.compute()
result = manager.get()[0]
assert len(result.policies) == 1
assert len(result.config.treatment_features) == 1
assert result.config.treatment_features[0] == 'AveRooms'
def test_causal_manager_global_cohort_policy(self, housing_data):
train_df, test_df, target_feature = housing_data
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, None)
manager.add(['AveRooms'])
manager.compute()
id = manager.get()[0].id
X_test = test_df.head(5).drop(target_feature, axis=1)
causal_data = manager.request_global_cohort_policy(id, X_test)
self.verify_common_causal_data_attributes(causal_data)
assert hasattr(causal_data, 'policies')
assert len(causal_data.policies[0].local_policies) == X_test.shape[0]
assert causal_data.policies[0].treatment_feature == "AveRooms"
incorrect_query_id = "incorrect_query_id"
X_test = test_df.drop(target_feature, axis=1)
with pytest.raises(ValueError,
match="Failed to find causal result with ID: "
"incorrect_query_id"):
manager.request_global_cohort_effects(incorrect_query_id, X_test)
def test_causal_cat_expansion(self, parks_data):
train_df, test_df, target_feature = parks_data
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, ['state', 'attraction'])
expected = "Increase the value 50"
with pytest.raises(ValueError, match=expected):
manager.add(['state'])
manager.compute()
def cost_manager(housing_data):
train_df, test_df, target_feature = housing_data
test_df = test_df[:7]
return CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, None)
def test_causal_manager_local_instance_effects(self, housing_data):
train_df, test_df, target_feature = housing_data
manager = CausalManager(train_df, test_df, target_feature,
ModelTask.REGRESSION, None)
manager.add(['AveRooms'])
manager.compute()
id = manager.get()[0].id
X_test = test_df.head(1).drop(target_feature, axis=1)
causal_data = manager.request_local_instance_effects(id, X_test)
self.verify_common_causal_data_attributes(causal_data)
assert hasattr(causal_data, 'local_effects')
EFFECTS_ATTRIBUTES = [
'sample',
'outcome',
'feature',
'feature_value',
'point',
'stderr',
'zstat',
'ci_lower',
'ci_upper',
'p_value',
]
for effect in EFFECTS_ATTRIBUTES:
assert effect in causal_data.local_effects[0][0]
incorrect_query_id = "incorrect_query_id"
X_test = test_df.drop(target_feature, axis=1)
with pytest.raises(ValueError,
match="Failed to find causal result with ID: "
"incorrect_query_id"):
manager.request_local_instance_effects(incorrect_query_id, X_test)
id = manager.get()[0].id
X_test = test_df.head(1).drop(target_feature, axis=1).values
with pytest.raises(UserConfigValidationException,
match='Data is of type <class \'numpy.ndarray\'>'
' but it must be a pandas DataFrame.'):
manager.request_local_instance_effects(id, X_test)
id = manager.get()[0].id
X_test = test_df.head(5).drop(target_feature, axis=1)
with pytest.raises(UserConfigValidationException,
match='Only one row of data is allowed for '
'local causal effects.'):
manager.request_local_instance_effects(id, X_test)