What is the right way to build wrappers around the test_train_split function with *args and **kwargs? To give more context, data science often require to create a test-valid