Overview

Objective

Teach practical aspects of productionizing ML services — from collecting requirements to model deployment and monitoring.

Target audience

Data scientists and ML engineers. Also software engineers and data engineers interested in learning about putting ML in production.

Pre-requisites

• Python
• Docker
• Being comfortable with command line
• Prior exposure to machine learning
• Prior programming experience (at least 1+ year)

Timeline

Course start: March 8th Course end: March 16th Final project presentation: March 31st

Syllabus

Module 1: Introduction

• What is MLOps
• Why do we need MLOps
• Running example: NY Taxi trips dataset
• Course overview
• Environment preparation
• Practice
• Homework

Module 2: Pipeline and Orchestration

• Introduction to workflows orchestration
• Introduction to Prefect
• From notebooks to workflows
• Continuous Training
• Next steps and resources
• Practice
• Homework

Module 3: Data Management

• Testing Data Quality
• Creating Data Pipelines
• Feature Stores
• Data Leakage
• Training / Serving Skew
• Practice
• Homework

Module 4: Model and Experiment Management

• Experiment tracking intro
• What is MLflow
• Experiment tracking with MLflow
• Saving and loading models with MLflow
• Model registry
• Practice
• Homework

Module 5: Model Deployment

• Web service: model deployment with FastAPI
• Docker: containerizing a web service
• MLflow: retrieving a model from the model registry
• Locust: load testing a web service

Module 6: Practice

• End-to-end project with all the things above

Instructors

• El Hamri Amale
• Guidi Ariel
• Quinquet Guillaume
• Si Larbi Karim
• Deschamps Mael