def test_cohort_filter_greater(self):
X_train, X_test, y_train, y_test, feature_names = create_iris_pandas()
filters = [{'arg': [2.8], 'column': SEPAL_WIDTH, 'method': 'greater'}]
validation_data = create_validation_data(X_test, y_test)
validation_data = validation_data.loc[X_test[SEPAL_WIDTH] > 2.8]
model_task = ModelTask.CLASSIFICATION
model = create_sklearn_svm_classifier(X_train, y_train)
categorical_features = []
run_rai_insights(validation_data,
model,
X_train,
y_train,
X_test,
y_test,
feature_names,
categorical_features,
model_task,
filters=filters)
def test_cohort_filter_index(self):
X_train, X_test, y_train, y_test, feature_names = create_iris_pandas()
# filter on index, which can be done from the RAI dashboard
filters = [{
'arg': [40],
'column': ROW_INDEX,
'method': 'less and equal'
}]
validation_data = create_validation_data(X_test, y_test)
validation_data = validation_data.loc[validation_data[ROW_INDEX] <= 40]
model_task = ModelTask.CLASSIFICATION
model = create_sklearn_svm_classifier(X_train, y_train)
categorical_features = []
model_task = ModelTask.CLASSIFICATION
run_rai_insights(validation_data,
model,
X_train,
y_train,
X_test,
y_test,
feature_names,
categorical_features,
model_task,
filters=filters)
def test_cohort_filter_multiclass_classification_outcome(
self, arg, correct_prediction):
X_train, X_test, y_train, y_test, feature_names = create_iris_pandas()
model = create_sklearn_svm_classifier(X_train, y_train)
model_task = ModelTask.CLASSIFICATION
categorical_features = []
# the index 1, corresponds to incorrect prediction
# the index 0 correspond to correct prediction
filters = [{
'arg': arg,
'column': CLASSIFICATION_OUTCOME,
'method': 'includes'
}]
pred_y = model.predict(X_test)
validation_data = create_validation_data(X_test, y_test, pred_y)
if correct_prediction:
validation_filter = validation_data[PRED_Y] == validation_data[
TRUE_Y]
else:
validation_filter = validation_data[PRED_Y] != validation_data[
TRUE_Y]
validation_data = validation_data.loc[validation_filter]
validation_data = validation_data.drop(columns=PRED_Y)
model_task = ModelTask.CLASSIFICATION
run_rai_insights(validation_data,
model,
X_train,
y_train,
X_test,
y_test,
feature_names,
categorical_features,
model_task,
filters=filters,
is_empty_validation_data=(not correct_prediction))