The interpretable appendix - using transparent modules to interpret opaque deep reinforcement learning agents