Ch 15: MLOps & Model Deployment - Intermediate¶
Track: Practitioner | Try code in Playground | Back to chapter overview
Read online or run locally
To run the code interactively, clone the repo and open chapters/chapter-15-mlops-and-model-deployment/notebooks/02_pipelines_cicd.ipynb in Jupyter.
Chapter 15: MLOps — Notebook 02 (Pipelines & CI/CD)¶
This notebook builds a reproducible sklearn Pipeline, sets up experiment tracking (MLflow with a JSON fallback), a file-backed model registry with stage transitions, and a GitHub Actions CI workflow that gates deploys on eval thresholds.
What you'll learn¶
| Topic | Section |
|---|---|
sklearn Pipeline for reproducible preprocessing + model | §1 |
| Reproducibility: seeds, lockfiles, the data/code/model triplet | §2 |
| Experiment tracking with MLflow (and JSON fallback) | §3 |
| File-backed model registry: stages and promotion gates | §4 |
| CI/CD with GitHub Actions: lint → test → train → eval → register → deploy | §5 |
| Quality gates and deploy approvals | §6 |
Time estimate: 2.5 hours
Key concepts¶
- sklearn Pipeline — Preprocess + model in one object — same transform at train and serve.
- Reproducibility triplet — Pin data version, code version, and model artifact together.
- Experiment tracking — Log params, metrics, and artifacts every run; compare runs without guesswork.
- Model registry — Stages (None / Staging / Production / Archived) gate which model serves traffic.
- CI eval gates — A run only promotes if metrics beat the previous Production model.
Run the full notebook for code and outputs.
Generated by Berta AI