https://airi.net/ru/articles/transformer-based-deep-reinforcement-learning-in-vizdoom/