Recurrent and multi-process PyTorch implementation of deep reinforcement Actor-Critic algorithms A2C and PPO