The Effect of Q-function Reuse on the Total Regret of Tabular, Model-Free, Reinforcement Learning.