Отправить по sms: Adaptive representations for reinforcement learning