Regression Tree Construction For Reinforcement Learning Problems With A General Action Space