Managing caching strategies for stream reasoning with reinforcement learning