Several code smell detection tools have been developed providing different results, because smells can be subjectively interpreted, and hence detected, in different ways. In this paper, we perform the largest experiment of applying machine learning algorithms to code smells to the best of our knowledge. We experiment 16 different machine-learning algorithms on four code smells (Data Class, Large Class, Feature Envy, Long Method) and 74 software systems, with 1986 manually validated code smell samples. We found that all algorithms achieved high performances in the cross-validation data set, yet the highest performances were obtained by J48 and Random Forest, while the worst performance were achieved by support vector machines. However, the lower prevalence of code smells, i.e., imbalanced data, in the entire data set caused varying performances that need to be addressed in the future studies. We conclude that the application of machine learning to the detection of these code smells can provide high accuracy (>96 %), and only a hundred training examples are needed to reach at least 95 % accuracy.

Comparing and experimenting machine learning techniques for code smell detection

Research on software testing produces many innovative automated techniques, but because software testing is by necessity incomplete and approximate, any new technique faces the challenge of an empirical assessment. In the past, we have demonstrated scientific advance in automated unit test generation with the EVOSUITE tool by evaluating it on manually selected open-source projects or examples that represent a particular problem addressed by the underlying technique. However, demonstrating scientific advance is not necessarily the same as demonstrating practical value; even if VOSUITE worked well on the software projects we selected for evaluation, it might not scale up to the complexity of real systems. Ideally, one would use large “real-world” software systems to minimize the threats to external validity when evaluating research tools. However, neither choosing such software systems nor applying research prototypes to them are trivial tasks. In this article we present the results of a large experiment in unit test generation using the VOSUITE tool on 100 randomly chosen open-source projects, the 10 most popular open-source projects according to the SourceForge Web site, seven industrial projects, and 11 automatically generated software projects. The study confirms that VOSUITE can achieve good levels of branch coverage (on average, 71p per class) in practice. However, the study also exemplifies how the choice of software systems for an empirical study can influence the results of the experiments, which can serve to inform researchers to make more conscious choices in the selection of software system subjects. Furthermore, our experiments demonstrate how practical limitations interfere with scientific advances, branch coverage on an unbiased sample is affected by predominant environmental dependencies. The surprisingly large effect of such practical engineering problems in unit testing will hopefully lead to a larger appreciation of work in this area, thus supporting transfer of knowledge from software testing research to practice.

/pdf/a-large-scale-evaluation-of-automated-unit-test-generation-24mhz66eca.pdf

A Large-Scale Evaluation of Automated Unit Test Generation Using EvoSuite

The field of big code relies on mining large corpora of code to perform some learning task. A significant threat to this approach has been recently identified by Lopes et al. (2017) who found a large amount of near-duplicate code on GitHub. However, the impact of code duplication has not been noticed by researchers devising machine learning models for source code. In this work, we explore the effects of code duplication on machine learning models showing that reported performance metrics are sometimes inflated by up to 100% when testing on duplicated code corpora compared to the performance on de-duplicated corpora which more accurately represent how machine learning models of code are used by software engineers. We present a duplication index for widely used datasets, list best practices for collecting code corpora and evaluating machine learning models on them. Finally, we release tools to help the community avoid this problem in future research.

The Adverse Effects of Code Duplication in Machine Learning Models of Code

Video games make up an important part of the software industry, yet the software engineering community rarely studies video games. This imbalance is a problem if video game development differs from general software development, as some game experts suggest. In this paper we describe a study with 14 interviewees and 364 survey respondents. The study elicited substantial differences between video game development and other software development. For example, in game development, “cowboy coders” are necessary to cope with the continuous interplay between creative desires and technical constraints. Consequently, game developers are hesitant to use automated testing because of these tests’ rapid obsolescence in the face of shifting creative desires of game designers. These differences between game and non-game development have implications for research, industry, and practice. For instance, as a starting point for impacting game development, researchers could create testing tools that enable game developers to create tests that assert flexible behavior with little up-front investment.

/pdf/cowboys-ankle-sprains-and-keepers-of-quality-how-is-video-7ew41sq3n9.pdf

Cowboys, ankle sprains, and keepers of quality: how is video game development different from software development?

Code smells are symptoms of poor design and implementation choices weighing heavily on the quality of produced source code. During the last decades several code smell detection tools have been proposed. However, the literature shows that the results of these tools can be subjective and are intrinsically tied to the nature and approach of the detection. In a recent work the use of Machine-Learning (ML) techniques for code smell detection has been proposed, possibly solving the issue of tool subjectivity giving to a learner the ability to discern between smelly and non-smelly source code elements. While this work opened a new perspective for code smell detection, it only considered the case where instances affected by a single type smell are contained in each dataset used to train and test the machine learners. In this work we replicate the study with a different dataset configuration containing instances of more than one type of smell. The results reveal that with this configuration the machine learning techniques reveal critical limitations in the state of the art which deserve further research.

Detecting code smells using machine learning techniques: Are we there yet?

In order to increase our ability to use measurement to support software development practise we need to do more analysis of code. However, empirical studies of code are expensive and their results are difficult to compare. We describe the Qualitas Corpus, a large curated collection of open source Java systems. The corpus reduces the cost of performing large empirical studies of code and supports comparison of measurements of the same artifacts. We discuss its design, organisation, and issues associated with its development.

/pdf/the-qualitas-corpus-a-curated-collection-of-java-code-for-24y62i28io.pdf

The Qualitas Corpus: A Curated Collection of Java Code for Empirical Studies

Empirical technology and society research is in a methodological crisis. Problems increasingly involve closed platforms, targeted content, and context-specific behavior. Prevailing research methods, such as surveys, tasks, and web crawls, pose design and ecological validity limitations. Deploying studies in participant browsers and devices is a promising direction. These vantage points can observe individualized experiences and implement UI interventions in real settings. We survey scholarship that uses these methods, annotating 284 sampled papers. Our analysis demonstrates their potential, but also recurring implementation barriers and shortcomings. We then present Rally and WebScience, a platform and toolkit for browser-based research. These systems lower implementation barriers and advance the science of measuring online behavior. Finally, we evaluate Rally and WebScience against our design goals. We report results from a one-month pilot study on news engagement, analyzing 4,466,200 webpage visits from 1,817 participants. We also present observations from interviews with researchers using these systems.

Ted Han

Papers

The Qualitas Corpus: A Curated Collection of Java Code for Empirical Studies

Rally and WebScience: A Platform and Toolkit for Browser-Based Research on Technology and Society Problems