Machine Learning (ML) · Advanced level

Advanced level · Module 5: MLOps and Monitoring · Lesson 1 of 4

Experiment Tracking and Versioning

Record every training run's data, code, parameters and results, and version models in a registry.

In this lesson you will

  • Track experiments with a tool such as MLflow
  • Version data and register models

Key concepts

experiment runparameters and metricsdata versioningmodel registry

The lesson

Without tracking, nobody remembers which settings produced the best model. Experiment tracking logs each run's parameters, metrics, code version, data version and artefacts automatically.

Version data as well as code: record a hash or version of the training data, so a model can always be traced to exactly what it learned from.

A model registry stores approved models with versions and stages such as staging and production, making it clear which model is live and allowing rollback.

Step by step

  1. Add tracking to the training script.
  2. Run several experiments and compare them in the tracking UI.
  3. Record the data version.
  4. Register the chosen model.

Practice activity

Track the forecasting experiments and register the production model.

Expected result: Every run is logged and the production model is registered with its data and code versions.

Check your understanding

  1. What does experiment tracking record?

    Show answer

    Parameters, metrics, code and data versions and artefacts of each run.

  2. What is a model registry for?

    Show answer

    Storing versioned, approved models and showing which is in production.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.