def test_rai_dashboard_input_adult_importances_success(
self, create_rai_insights_object_classification):
ri = create_rai_insights_object_classification
dashboard_input = ResponsibleAIDashboardInput(ri)
flask_server_prediction_output = dashboard_input.importances()
self.check_success_criteria(flask_server_prediction_output)
def test_rai_dashboard_input_housing_causal_global_policy_success(
self, create_rai_insights_object_regression, filters):
ri = create_rai_insights_object_regression
dashboard_input = ResponsibleAIDashboardInput(ri)
id = ri.causal.get()[0].id
post_data = (id, filters, [])
flask_server_prediction_output = \
dashboard_input.get_global_causal_policy(
post_data)
self.check_success_criteria(flask_server_prediction_output)
def test_rai_dashboard_input_adult_on_predict_failure(
self, create_rai_insights_object_classification):
ri = create_rai_insights_object_classification
test_data = ri.test
dashboard_input = ResponsibleAIDashboardInput(ri)
test_pred_data = test_data.head(1).values
flask_server_prediction_output = dashboard_input.on_predict(
test_pred_data)
self.check_failure_criteria(
flask_server_prediction_output,
"Model threw exception while predicting...")
def test_model_analysis_adult(self,
create_rai_insights_object_classification):
ri = create_rai_insights_object_classification
knn = ri.model
test_data = ri.test
dashboard_input = ResponsibleAIDashboardInput(ri)
with patch.object(knn, "predict_proba") as predict_mock:
test_pred_data = test_data.head(1).drop("Income", axis=1).values
dashboard_input.on_predict(test_pred_data)
assert (
predict_mock.call_args[0][0].values == test_pred_data).all()
def test_rai_dashboard_input_housing_causal_global_effects_failure(
self, create_rai_insights_object_regression):
ri = create_rai_insights_object_regression
dashboard_input = ResponsibleAIDashboardInput(ri)
id = "some_id_that_does_not_exist"
post_data = (id, [], [])
flask_server_prediction_output = \
dashboard_input.get_global_causal_effects(
post_data)
self.check_failure_criteria(
flask_server_prediction_output,
"Failed to generate global causal effects for cohort,")
def test_rai_dashboard_input_iris_on_predict_success(
self, create_rai_insights_object_multiclass_classification):
ri = create_rai_insights_object_multiclass_classification
rf = ri.model
test_data = ri.test
dashboard_input = ResponsibleAIDashboardInput(ri)
test_pred_data = test_data.head(1).drop("target", axis=1).values
flask_server_prediction_output = dashboard_input.on_predict(
test_pred_data)
rf_prediction = rf.predict_proba(test_pred_data)
assert rf_prediction is not None
assert (flask_server_prediction_output['data'] == rf_prediction).all()
self.check_success_criteria(flask_server_prediction_output)
def test_rai_dashboard_input_housing_causal_whatif_success(
self, create_rai_insights_object_regression):
ri = create_rai_insights_object_regression
id = ri.causal.get()[0].id
causal_whatif_test_data = ri.test.head(1).drop(
"target", axis=1).to_dict(orient='records')
treatment_feature = 'AveRooms'
current_treatment_value = [
causal_whatif_test_data[0][treatment_feature]
]
current_outcome = [ri.test.head(1)["target"].values[0]]
dashboard_input = ResponsibleAIDashboardInput(ri)
post_data = (id, causal_whatif_test_data, treatment_feature,
current_treatment_value, current_outcome)
flask_server_prediction_output = dashboard_input.causal_whatif(
post_data)
self.check_success_criteria(flask_server_prediction_output)
def test_rai_dashboard_input_adult_debug_ml_success(
self, create_rai_insights_object_classification):
ri = create_rai_insights_object_classification
features = ri.test.drop("Income", axis=1).columns.tolist()
filters = []
composite_filters = []
max_depth = 3
num_leaves = 3
min_child_samples = 8
metric = "Error rate"
post_data = [
features, filters, composite_filters, max_depth, num_leaves,
min_child_samples, metric
]
dashboard_input = ResponsibleAIDashboardInput(ri)
flask_server_prediction_output = dashboard_input.debug_ml(post_data)
self.check_success_criteria(flask_server_prediction_output)
def test_rai_dashboard_input_adult_matrix_failure(
self, create_rai_insights_object_classification):
ri = create_rai_insights_object_classification
features = ['Age', 'Workclass']
filters = []
composite_filters = []
quantile_binning = False
num_bins = 8
metric = "Error Rate"
post_data = [
features, filters, composite_filters, quantile_binning, num_bins,
metric
]
dashboard_input = ResponsibleAIDashboardInput(ri)
flask_server_prediction_output = dashboard_input.matrix(post_data)
self.check_failure_criteria(
flask_server_prediction_output,
"Failed to generate json matrix representation,")
def test_rai_dashboard_input_adult_matrix_success(
self, create_rai_insights_object_classification):
ri = create_rai_insights_object_classification
features = ['Age', 'Workclass']
filters = []
composite_filters = []
quantile_binning = False
num_bins = 8
metric = "Error rate"
post_data = [
features, filters, composite_filters, quantile_binning, num_bins,
metric
]
dashboard_input = ResponsibleAIDashboardInput(ri)
flask_server_prediction_output = dashboard_input.matrix(post_data)
self.check_success_criteria(flask_server_prediction_output)
empty_features = [None, None]
post_data = [
empty_features, filters, composite_filters, quantile_binning,
num_bins, metric
]
dashboard_input = ResponsibleAIDashboardInput(ri)
flask_server_prediction_output = dashboard_input.matrix(post_data)
assert len(flask_server_prediction_output[
WidgetRequestResponseConstants.data]) == 0
self.check_success_criteria(flask_server_prediction_output)
def __init__(self,
analysis: RAIInsights,
public_ip=None,
port=None,
locale=None,
cohort_list=None,
**kwargs):
self.input = ResponsibleAIDashboardInput(analysis,
cohort_list=cohort_list)
super(ResponsibleAIDashboard,
self).__init__(dashboard_type="ResponsibleAI",
model_data=self.input.dashboard_input,
public_ip=public_ip,
port=port,
locale=locale,
no_inline_dashboard=True,
**kwargs)
def predict():
data = request.get_json(force=True)
return jsonify(self.input.on_predict(data))
self.add_url_rule(predict, '/predict', methods=["POST"])
def tree():
data = request.get_json(force=True)
return jsonify(self.input.debug_ml(data))
self.add_url_rule(tree, '/tree', methods=["POST"])
def matrix():
data = request.get_json(force=True)
return jsonify(self.input.matrix(data))
self.add_url_rule(matrix, '/matrix', methods=["POST"])
def causal_whatif():
data = request.get_json(force=True)
return jsonify(self.input.causal_whatif(data))
self.add_url_rule(causal_whatif, '/causal_whatif', methods=["POST"])
def global_causal_effects():
data = request.get_json(force=True)
return jsonify(self.input.get_global_causal_effects(data))
self.add_url_rule(global_causal_effects,
'/global_causal_effects',
methods=["POST"])
def global_causal_policy():
data = request.get_json(force=True)
return jsonify(self.input.get_global_causal_policy(data))
self.add_url_rule(global_causal_policy,
'/global_causal_policy',
methods=["POST"])
def importances():
return jsonify(self.input.importances())
self.add_url_rule(importances, '/importances', methods=["POST"])
class ResponsibleAIDashboard(Dashboard):
"""The dashboard class, wraps the dashboard component.
:param analysis: An object that represents an RAIInsights.
:type analysis: RAIInsights
:param public_ip: Optional. If running on a remote vm,
the external public ip address of the VM.
:type public_ip: str
:param port: The port to use on locally hosted service.
:type port: int
:param locale: The language in which user wants to load and access the
ResponsibleAI Dashboard. The default language is english ("en").
:type locale: str
:param cohort_list:
List of cohorts defined by the user for the dashboard.
:type cohort_list: List[Cohort]
"""
def __init__(self,
analysis: RAIInsights,
public_ip=None,
port=None,
locale=None,
cohort_list=None,
**kwargs):
self.input = ResponsibleAIDashboardInput(analysis,
cohort_list=cohort_list)
super(ResponsibleAIDashboard,
self).__init__(dashboard_type="ResponsibleAI",
model_data=self.input.dashboard_input,
public_ip=public_ip,
port=port,
locale=locale,
no_inline_dashboard=True,
**kwargs)
def predict():
data = request.get_json(force=True)
return jsonify(self.input.on_predict(data))
self.add_url_rule(predict, '/predict', methods=["POST"])
def tree():
data = request.get_json(force=True)
return jsonify(self.input.debug_ml(data))
self.add_url_rule(tree, '/tree', methods=["POST"])
def matrix():
data = request.get_json(force=True)
return jsonify(self.input.matrix(data))
self.add_url_rule(matrix, '/matrix', methods=["POST"])
def causal_whatif():
data = request.get_json(force=True)
return jsonify(self.input.causal_whatif(data))
self.add_url_rule(causal_whatif, '/causal_whatif', methods=["POST"])
def global_causal_effects():
data = request.get_json(force=True)
return jsonify(self.input.get_global_causal_effects(data))
self.add_url_rule(global_causal_effects,
'/global_causal_effects',
methods=["POST"])
def global_causal_policy():
data = request.get_json(force=True)
return jsonify(self.input.get_global_causal_policy(data))
self.add_url_rule(global_causal_policy,
'/global_causal_policy',
methods=["POST"])
def importances():
return jsonify(self.input.importances())
self.add_url_rule(importances, '/importances', methods=["POST"])
