Text this: Improving the Efficiency of Deep Reinforcement Learning in NetHack Using VAE and Proposing Additional Rewards.