Introducing MLOps to Facilitate the Development of Machine Learning Models in Agronomy: A Case Study

While machine learning (ML) and deep learning (DL) are increasingly being adopted in agronomy, the literature shows that the use of ML Operations (MLOps) frameworks remains scarce during the research stage. This study discusses the utility of MLOps in enhancing the reproducibility and transparency of ML projects even before deployment in production. Moreover, by showing the process of integrating these practices in a pre-existing codebase from an ML study in agronomy, this work showcases the transformative potential of MLOps and serves as a guideline for efficiently utilizing these frameworks. Among the many MLOps frameworks, ClearML was selected and integrated into all the project’s aspects, from data processing through hyper-parameter optimization to orchestrating the entire workflow. This integration not only streamlined dataset management, experiment tracking, and hyper-parameter optimization but also demonstrated the utility of a more structured approach to ML project design, leading to more informed design choices. Our findings underscore the potential of these practices to revolutionize the reproducibility, transparency, and design of ML projects in computational agronomy research.