On the value of combining feature subset selection with genetic algorithms
James H. Andrews, Tim Menzies · 2009
The next challenge for the PROMISE community is scaling up and speeding up model generation to meet the size and time constraints of modern software development projects. There will always be a trade-off between completeness and runtime speed. Here we explore that trade-off in the context of using genetic algorithms to learn coverage models; i.e. biases in the control structures for randomized test generators. After applying feature subset selection to logs of the GA output, we find we can generate the coverage model and run the resulting test suite ten times faster while only losing 6% of the test case coverage.