The Impact on Time Efficiency of Varying Mutation Rate and Population Size in jGenProg

Sammanfattning

Automated Program Repair (APR) is an approach to automate the debugging task in software development and to make it more efficient. However, reaching a solution can still be time consuming. jGenProg is a tool in APR using genetic programming in its repair approach, where the parameter values “mutation rate” and “population size” are used to change the program code into potential solutions (called program variants) and to determine how many program variants to evaluate. The default values of mutation rate and population size are both set to 1. Previous studies have shown that different values of mutation rate and population size can affect the efficiency of genetic programming. This has, to our knowledge, not been studied in genetic programming-based APR tools like jGenProg. In our empirical study, we conducted an experiment exploring how time efficiency is affected by varying the values of mutation rate and population size in jGenProg. We tested a subset of bugs from a widely used dataset called Defects4J and measured the average runtime upon reaching a solution. The results showed that the default values of mutation rate and population size had the shortest average runtime of 16 minutes, and the second highest number of successful instances. Our conclusion is that time efficiency is affected by the values of mutation rate and population size, where the most time-efficient combination was the default values of mutation rate and population size. An increasing population size almost always led to an increase in average runtime, making the process less time-efficient with larger population sizes.