Reinforcement Learning with Action-Specific Focuses in Video Games