Text this: Nonstationary Stochastic Bandits: UCB Policies and Minimax Regret