Reinforced Domain Selection for Continuous Domain Adaptation