Where’s the theory on datasets? The canonical model of splitting a dataset into train and test data allows statistical validity for the evaluation of model performance. [...]