Describir: Deep variational reinforcement learning for POMDPs