Sample and memory efficiency in episodic reinforcement learning