R3L: Connecting Deep Reinforcement Learning To Recurrent Neural ...