Machine Learning (ML) · Advanced level

Advanced level · Module 5: MLOps and Monitoring · Lesson 3 of 4

Retraining, Rollback and Governance

Decide when and how to retrain, roll back safely and assign clear responsibility for models.

In this lesson you will

  • Define retraining triggers and a safe release process
  • Assign ownership and governance for a model

Key concepts

retraining triggersshadow and champion–challengerrollbackmodel ownership

The lesson

Retrain on a schedule, when monitoring triggers an alert, or both. Retrained models must pass the same evaluation and fairness checks before release.

Release carefully: run a new model in shadow mode beside the current one, or compare champion and challenger on recent data, then promote it. Keep the previous version ready for rollback.

Every production model needs a named owner, a review schedule and documented decisions, including when to switch it off.

Step by step

  1. Define retraining triggers for the forecaster.
  2. Retrain on data including the shock.
  3. Compare champion and challenger.
  4. Document the release or rollback decision and owner.

Practice activity

Complete the drift response from the practical activity with a documented retraining decision.

Expected result: A documented decision with evaluation evidence, release steps and a named owner.

Check your understanding

  1. What is shadow mode?

    Show answer

    Running a new model alongside the current one without using its outputs, to compare them.

  2. Why keep the previous model version?

    Show answer

    To roll back quickly if the new one performs worse.

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.