Model Deployment & MLOps
Machine Learning · Advanced
Definition
Deployment serves a trained model behind an API; MLOps automates training, versioning and monitoring.
A model only creates value once it is used in production.
Diagram
Flowchart
Example: Drift check
train_mean, live_mean, std = 50, 58, 5
print('drift' if abs(live_mean - train_mean) > std else 'ok')Output: drift
Key points
- Version data, code and models
- Log predictions for audits
Interview questions
What is data drift?
When live input data no longer looks like training data.