Ensuring the reproducibility and traceability of experiments is a key challenge in conducting machine learning research projects. Tools derived from software engineering and industrial machine learning provide an effective solution to these needs thanks to their low integration costs and high utility.

This workshop offers a hands-on introduction to four tools widely used in modern MLOps (Machine Learning Operations) workflows that are well-suited for research projects in Python:

• uv for environment and dependency management.
• Hydra for modular experiment configuration.
• MLflow for tracking and tracing experiments.

Through guided development of a simplified project, workshop participants had the opportunity to gain a deep understanding of the tools’ philosophy, structure an experimental pipeline (processing and version management of datasets, model training, basic hyperparameter optimization). The workshop aims to highlight how these tools facilitate code organization, the reproducibility of experiments, and collaboration within research teams.

I conducted this workshop in collaboration with Joseph Allyndrée for the State of The R working group.

Resources

• Slides

• GitHub repository for the practical component