MLOps and production deployment

Objectives

• Version data, code and models, and retrieve any version of a model together with its parameters and metrics.
• Package a model into a containerised API, test it and document it.
• Automate training, testing and deployment in a continuous-integration pipeline triggered by a change of code or data.
• Deploy the model on a managed platform (SageMaker, Azure ML or Google AI Platform, depending on the organisation's constraint) or on a Kubernetes cluster.
• Set up monitoring of data drift and performance, with alert thresholds and a written retraining procedure.

Audience

Data and machine learning engineers, software and DevOps engineers responsible for models, and IT teams who take over a model in production.

Prerequisites

Having completed DATA-201, or having a model trained in Python, and using Git. Some knowledge of Docker is useful but not required. Checked by a placement test. One laptop per participant, 8 GB of memory minimum, with Python 3, Docker, Git and a code editor installed before the first day; training cloud accounts, credits and a training Git repository provided by Yolea; for remote delivery, a stable connection of at least 10 Mbit/s, microphone and camera. A training model is provided; the participant's own model is accepted if it is ready to deploy.

Programme

• Model version management
• Exposing a model as an API
• Continuous integration and deployment
• Cloud or Kubernetes deployment
• Monitoring and maintaining a model in production

Format

In-company or open enrolment. 21 hours, 3 consecutive days of 7 hours, or 2 + 1 days one week apart with coaching between the modules. On site recommended, hybrid possible; in-company on your own platform, or open enrolment; 8 to 12 participants. 25% input, 75% supervised building; one running model followed from the Git repository to the monitoring dashboard. Tools used: MLflow for tracking, FastAPI and Docker for serving, GitHub Actions for the pipeline, Kubernetes or your own managed platform for deployment.

Assessment

Beforehand, a placement test. During the course, check quizzes and a pipeline review. At the end, a practical challenge (deploy an imposed model from end to end; assessed against a grid: versioning, tests, deployment, monitoring), a same-day feedback survey and a named certificate of course completion. Three weeks later, a follow-up check-in.