It abstracts away the complexity of the inference and training portions of the RL loop while allowing for some custom configuration. An outline of the training loop is shown below: Inference Your code ...