Dynamic scope discovery for model transformations

Abstract

Optimizations to local-search based model transformations typically aim at effectively ordering the traversal of pattern edges to reduce the search space. In this paper we propose a dynamic approach to on-line discovery of rule application areas. Our approach incorporates tracking transformation progress in the input model using temperature-based coloring of model elements. The resulting heat map is used to discover possible rule application scopes ahead of rule execution. Further refinement of scopes is achieved by applying a Naive Bayes (NB) classifier to predict a set of possible match candidates. NB is well suited for the computationally intensive environment of model transformations due to its incremental training phase and low classification overhead. Our design is intended to take a runtime, black-box approach to observing and learning from the transformations as they are executed. Finally, we demonstrate a prototype evaluation of the approach in our transformation tool AToMPM [24] and address the benefits, limitations as well as future applications.