Experiment Tracking for MLOps Engineers
Experiment tracking records every training run with its parameters, metrics, code version and artefacts so results can be compared and reproduced. MLflow and Weights and Biases are common tools, and a model registry stores versioned models with stages. This unit sits in the Data and Experimentation stage.
Stage: Data and Experimentation · Part of the MLOps Engineer roadmap · Last reviewed 19 November 2026
What to learn in Experiment Tracking
- Log parameters, metrics and artefacts from a training script using MLflow
- Compare runs in a tracking UI to choose the best configuration
- Register a model version and move it through staging to production
- Tag runs with the Git commit and dataset version
- Set a random seed and record the environment to make runs repeatable
Experiment Tracking interview questions
What does a model registry add beyond experiment tracking?
Tracking records runs. A registry manages the chosen models as versioned releases with metadata, lineage and lifecycle stages or aliases such as staging and production, and often approval steps. Deployment systems then pull a specific registered version.
What should you log for every experiment?
Log the code commit, dataset version, hyperparameters, random seed, environment or dependency versions, evaluation metrics and the produced model artefact. With these, anyone can rerun the experiment and explain differences between two runs.
Where Experiment Tracking fits in the MLOps Engineer roadmap
Experiment Tracking belongs to the Data and Experimentation stage (intermediate level), which takes about 4 hours. Reproducible science.
Other units in this stage
Back to the full MLOps Engineer roadmap