Test set verification is an essential step in model building

Abstract

Recently, Christin et al. published an article that reviewed the field of deep learning and offered advice on how to train a deep learning model. We write here to emphasize the importance of model verification, which can help ensure that the model will generalize to new data. Specifically, we discuss the importance of using a test set for model verification, and of defining an explicit research hypothesis. We then present a revised workflow that will help ensure that the accuracy reported for your deep learning model is reliable.