Optimization Of TCP Reinforcement Learning Method For Continous Integeration

Continuous Integration(CI)requires rapid software construction and testing.When the test suite becomes larger and larger with the software update iteration,the test execution process will become slower and more inefficient.The core of CI is to achieve rapid construction and feedback.Correspondingly,regression testing is required to be completed within a limited time.The use of test case prioritization(TCP)and other test optimization techniques can effectively reduce the test time in CI development.Existing research shows that the nature of the TCP problem for CI is a sequential decision-making problem,which can be solved by reinforcement learning,and the reward mechanism is an important part of the construction of a reinforcement learning system,and it is also an important factor to determine the learning efficiency of the reinforcement learning agent in the continuous integration testing.This paper conducts research on the reinforcement learning reward mechanism for continuous integration testing,including the design of reward function and the selection of reward objects,in order to optimize the continuous integration testing process.In the design of the reward function,aiming at the problem of excessive historical information,by analyzing the historical failure information of the test case,the failure interval of the historical failure information of the test case is studied,and a reward function based on dynamic time window,named APHF DTW,is proposed.In the selection of reward objects,for a single strategy that only rewards failed test cases,the test case similarity metric method based on test case history information sequence and test time is defined,and a reinforcement learning reward strategy based on test case similarity is further proposed.This paper conducts empirical research on 4 research questions on 6 industrial data sets to verify the effectiveness of the reward mechanism proposed in this paper.The experimental results show that(1)The reward function based on dynamic time window can effectively improve the learning efficiency of the reinforcement learning agent,thereby improving the prioritization effect of the test sequence.(2)Under the guidance of the reward strategy based on test case similarity,the performance of the reward function based on dynamic time window has been further improved.(3)The reward mechanism proposed in this paper will increase the overall integration test time due to the calculation of the dynamic time window and the similarity value between the test case,but the time increase is in the second range,for the test of the actual industrial level program,the time increase in the second range is acceptable.