Data-Driven Regret Balancing for Online Model Selection in Bandits