def test_explain_model_imbalanced_classes(self, mimic_explainer):
model = retrieve_model('unbalanced_model.pkl')
x_train = retrieve_dataset('unbalanced_dataset.npz')
model_predictions = model.predict(x_train)
# Assert the model's predictions are skewed
assert len(np.unique(model_predictions)) == 2
explainable_model = LGBMExplainableModel
explainer = mimic_explainer(model, x_train, explainable_model, max_num_of_augmentations=10)
global_explanation = explainer.explain_global(x_train, include_local=True)
# There should be an explanation per feature
assert len(global_explanation.global_importance_values) == 1585
# We should get back an explanation for each class
assert len(global_explanation.local_importance_values) == 3
# Get the underlying multiclass model
surrogate_predictions = explainer.surrogate_model.model.predict(x_train)
assert len(np.unique(surrogate_predictions)) == 2
assert len(np.unique(model_predictions)) == 2
assert np.isclose(surrogate_predictions, model_predictions).all()
def preload_models():
for model_name in MODELS_NAMES:
retrieve_model(model_name)