Adaptive Reward Computation in Reinforcement Learning-Based Continuous Integration Testing

摘要: Reinforcement learning (RL) has been applied to prioritizing test cases in Continuous Integration (CI) testing, where the reward plays a crucial role. It demonstrated that historical information-based function can improve effectiveness of case prioritization (TCP). However, inherent character frequent iterations CI produce considerable accumulation information, which may decrease TCP efficiency and result slow feedback. In this paper, partial information is considered computation, sliding window techniques are adopted capture possible efficient information. Firstly, fixed-size introduced set fixed length recent for each test. Then dynamic proposed, size continuously adaptive testing. Two methods suite-based individual case-based window. The empirical studies conducted on fourteen industrial-level programs, results reveal under limited time, window-based effectively effect, NAPFD (Normalized Average Percentage Faults Detected) Recall windows better than particular, approach rank 74.18% failed top 50% sorting sequence, with 1.35% improvement 6.66 positions increased TTF (Test Fail).