Compressing Uniform Test Suites Using Variational Autoencoders
André Reichstaller, Alexander Knapp · 2017
Uniform test suites consist of test cases exclusively differing in test inputs - not in test goals. Intended to gain confidence that a given invariant holds, these inputs trigger particular behavior of the system under test. Equipped with a simulation of the system under test we are able to cheaply explore this behavior virtually. When changing over to reality, testing the system within its real context, we are, however, often forced to limit the number of test inputs.We consider the task of compressing such uniform test suites. Similar to the task of minimizing classical test suites, this activity aims at finding a small number of representatives out of many test cases that are composed in the suite. In contrast, for uniform test suites we only have to take test inputs into account. As we know the uniform test goal, i.e., validating the invariants, we are able to automatically derive even new test cases within the compression. This work reports on first experiments with using variational autoencoders for this task. We show that generating test cases by utilizing these neural models might provide the tester with completely new insights while still performing well w.r.t. a presented compression coverage goal.