On Bonus-Based Exploration Methods in the Arcade Learning Environment