Design Space Exploration of Approximate Computing Techniques with a Reinforcement Learning Approach