An Improved Reinforcement Learning Algorithm for Learning to Branch