Towards a stream based machine learning testbed

With the rise of data analytics in industrial applications a heterogeneous tool landscape developed over the past few years. To cope with highly dynamic and domain-specific requirements of such applications, scripting programming languages and frameworks, which offer ecosystems comprising numerous publicly available plugin libraries, are gaining more and more attraction as a starting point, since they enable rapid prototyping and thus, quick results. At the other end, cloud service providers are continuously extending their data analysis product palette in order to support large enterprise solutions for real-world deployments. As scripted prototypes and cloud based solutions have their strengths in different phases of an analytics project, we identified several pitfalls in recent case studies when moving a prototypic approach to release. In this work, we present a software design for a data stream analysis testbed, with the aim to address some of these challenges. Therefore, an interaction pattern for common analysis steps (data acquisition, visualization, preprocessing and machine learning model evaluation) is detailed, results gained from a sample case study are summarized and future leads are discussed.