Pinset: A DSL for Extracting Datasets from Models for Data Mining-Based Quality Analysis

Abstract

Data mining techniques have been successfully applied to software quality analysis and assurance, including quality of modeling artefacts. Before such techniques can be used, though, data under analysis commonly need to be formatted into two-dimensional tables. This constraint is imposed by data mining algorithms, which typically require a collection of records as input for their computations. The process of extracting data from the corresponding sources and formatting them properly can become error-prone and cumbersome. In the case of models, this process is mostly carried out through scripts written in a model management language, such as EOL or ATL. To improve this situation, we present Pinset, a domain-specific language devised for the extraction of tabular datasets from software models. Pinset offers a tailored syntax and built-in facilities for common activities in dataset extraction. For evaluation, Pinset has been used on UML class diagrams to calculate metrics that can be employed as input for several fault-prediction algorithms. The use of Pinset for this calculations led to more compact and highlevel specifications when compared to equivalent scripts written in generic model management languages.