Detecting drift in deep learning: A methodology primer