A flexible and expressive formalism to specify metamorphic properties for BIG DATA systems validation

Abstract

BIG DATA systems represent a huge challenge for software engineering validations tasks since they have been classified as "non testable". Metamorphic Relationships (MR) have been proposed as a technique to overcome this problem. These relationships establish interactions between data that can be used to validate the expected behavior of the system. However, the process of exploring and defining MRs is a very arduous one, and an expressive and flexible specification language is needed to denote them. In this work we show how the Feather Weight Visual Scenarios (FVS) framework can be seen as an appealing tool to specify MRs. We exploit FVS features to model complex MR interactions and analysis, allowing the possibility to perform non trivial operations between MRs such as refinement and consistency checking. FVS is shown in action by introducing a proof of concept example focused on a machine learning system over biology cell images.