Training Trains Internship
Student login
← MLOps learning path
MLOps · Module 2 of 10

Reproducible experiments

Track code, data, parameters and results.

📖 Key ideas

  1. Version code with Git.
  2. Version data and models with DVC or lakeFS.
  3. Track experiments (params, metrics, artifacts) with MLflow or W&B.
  4. Pin dependencies (requirements.txt, conda env, Docker).
  5. Set random seeds.
💡 Real-life example: Experiment tracking is a lab notebook recording every ingredient and result, so you can repeat the best one.
DVCData Version Control
MLflow TrackingLogs params, metrics and artifacts
ArtifactOutput file like a model
Pinned dependenciesExact package versions