Steffen Tunkel

Steffen Herbold, Steffen Tunkel

Context: Differential testing is a useful approach that uses different implementations of the same algorithms and compares the results for software testing. In recent years, this approach was successfully used for test campaigns of deep learning frameworks. Objective: There is little knowledge on the application of differential testing beyond deep learning. Within this article, we want to close this gap for classification algorithms. Method: We conduct a case study using Scikit-learn, Weka, Spark MLlib, and Caret in which we identify the potential of differential testing by considering which algorithms are available in multiple frameworks, the feasibility by identifying pairs of algorithms that should exhibit the same behavior, and the effectiveness by executing tests for the identified pairs and analyzing the deviations. Results: While we found a large potential for popular algorithms, the feasibility seems limited because often it is not possible to determine configurations that are the same in other frameworks. The execution of the feasible tests revealed that there is a large amount of deviations for the scores and classes. Only a lenient approach based on statistical significance of classes does not lead to a huge amount of test failures. Conclusions: The potential of differential testing beyond deep learning seems limited for research into the quality of machine learning libraries. Practitioners may still use the approach if they have deep knowledge about implementations, especially if a coarse oracle that only considers significant differences of classes is sufficient.

Steffen Tunkel, Steffen Herbold

Context: Performance metrics are a core component of the evaluation of any machine learning model and used to compare models and estimate their usefulness. Recent work started to question the validity of many performance metrics for this purpose in the context of software defect prediction. Objective: Within this study, we explore the relationship between performance metrics and the cost saving potential of defect prediction models. We study whether performance metrics are suitable proxies to evaluate the cost saving capabilities and derive a theory for the relationship between performance metrics and cost saving potential. Methods: We measure performance metrics and cost saving potential in defect prediction experiments. We use a multinomial logit model, decision, and random forest to model the relationship between the metrics and the cost savings. Results: We could not find a stable relationship between cost savings and performance metrics. We attribute the lack of the relationship to the inability of performance metrics to account for the property that a small proportion of very large software artifacts are the main driver of the costs. Conclusion: Any defect prediction study interested in finding the best prediction model, must consider cost savings directly, because no reasonable claims regarding the economic benefits of defect prediction can be made otherwise.