Chapter 15: MLOps & Model Deployment¶

Take a model from notebook to production—package with joblib, serve with FastAPI, containerize, build a model registry, design CI/CD with eval gates, and monitor drift, latency, and errors in production.

Metadata¶

Learning Objectives¶

• Package ML models for production: serialize sklearn pipelines, freeze deps, define typed I/O
• Serve models behind an HTTP API with FastAPI (/predict, /health, /version, batching)
• Containerize and reason about deployments: Dockerfile layers, image size, health/readiness probes
• Track experiments and manage a model registry with stages and promotion gates
• Design CI/CD for ML: lint → test → train → eval → register → deploy with quality gates
• Monitor models in production: PSI / KS drift, latency, errors, A/B tests, canary, rollback

What's Included¶

Notebooks¶

Scripts¶

• config.py — Chapter config, registry paths, optional-integration flags
• deployment.py — FastAPI service factory, Pydantic schemas, batching helpers
• registry.py — File-backed model registry with stage transitions (None / Staging / Production / Archived)
• monitoring.py — Drift (PSI, KS), latency percentiles, structured JSON logs

Exercises¶

• Problem Set 1 (notebook) — Package a model with joblib, write a Pydantic schema, build /predict, write a Dockerfile, batch predictions, add /version
• Problem Set 2 (notebook) — Detect drift via PSI, implement a canary splitter, write a CI YAML with eval gates, build a tiny registry, structured logging middleware, rollback policy
• Solutions — In exercises/solutions/ (notebooks and solutions.py for CI)

Diagrams (Mermaid)¶

• mlops_lifecycle.mermaid, deployment_architecture.mermaid, monitoring_pipeline.mermaid

Read Online¶

• 15.1 Introduction — Lifecycle, joblib, Pydantic, FastAPI, Dockerfile, health checks
• 15.2 Intermediate — Pipelines, reproducibility, tracking, registry, GitHub Actions CI
• 15.3 Advanced — Drift detection, A/B & canary, observability, scaling, capstone

Or try the code in the Playground.

How to Use This Chapter¶

1. Clone and install dependencies

git clone https://github.com/luigipascal/berta-chapters.git
cd berta-chapters
pip install -r requirements.txt

2. Navigate to the chapter

cd chapters/chapter-15-mlops-and-model-deployment
pip install -r requirements.txt

3. (Optional) Install MLflow / Evidently / Prometheus

pip install mlflow evidently prometheus-client bentoml
# All notebooks fall back to local implementations if these are missing.

4. Launch Jupyter

jupyter notebook notebooks/01_packaging_serving.ipynb

Created by Luigi Pascal Rondanini | Generated by Berta AI