They can be quite difficult to configure and apply to arbitrary sequence prediction problems, even with well defined and “easy to use” interfaces like those provided in the Keras deep learning library in Python. One reason for this difficulty in Keras is the use of the TimeDistributed wrapper...