DevOps Practices Into Machine Learning

All industrial machine learning (ML) projects have their ultimate objective to quickly develop and deploy ML solutions. However, a lot of machine learning projects are failing, and never reach production. In order to address these concerns, this paper aims to investigate how manual ML processes can be automated and operationalised so that more ML concept can be put into production. The advantages of DevOps automation and machine learning capabilities are now being combined by industries. MLOps is the name of this combination. Our analysis found that while there are some similar aspects in defining MLOps, still there is a lack of standardisation in the term, which can occasionally lead to misunderstandings and uncertainty. We carried out the research, which includes a review of the literature and an evaluation of the MLOps tools, to fill this gap. These investigations have led us to define MLOps and identify unresolved issues in the field. In addition to this, we offer a workflow-based summary of all essential components. Finally, this paper gives guidance for applying DevOps practises to machine learning for researchers and practitioners.