Co-adaptation of algorithmic and implementational innovations in inference-based deep reinforcement learning