I recently joined the MLOps zoomcamp, it is a 3-months course that goes through the basic foundations of MLOps.
We began week 1 recently, and would love to record my learning process and share my knowledge publicly.
Notes:
MLOps is a set of best practices for putting machine learning in production.
The machine learning cycle involves 3 main stages: Design, Train, Operate.
As an organization matures there is need to have a structured way to track experiments, register models, avoid redundancies using ML Pipelines.
As ML needs grows, an organization's level of maturity of MLOps can be categorized under 5 levels, numbering starts from zero.
0. No MLOps
No automation
All code in jupyter
1. DevOps, No MLOps
Releases are automated
Unit and Integration Tests
CI/CD
OPS Metrics
No experiment tracking
No reproducibility
DS Separated from engineers
2. Automated Training
Training pipeline
Experiment tracking
Model Registry
Low Friction Deployment
DS Works with Engineer
3. Automated Deployment
Easy to deploy model
A/B Test
4. Full MLOps Automation
Top comments (0)