AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsMedium
A large e-commerce company uses a machine learning model to personalize product recommendations. The model is deployed on a SageMaker real-time endpoint. The data science team frequently updates the model (e.g., weekly) and wants to introduce new model versions to a small subset of users (e.g., 5%) before rolling it out to all users. They need to monitor the performance of the new model version against the current production model without impacting the majority of users. Which deployment strategy should they employ?
- AShadow Deployment.
- BA/B Testing.
- CBlue/Green Deployment.
- DCanary Deployment.
Show answer & explanationAnswer & explanation
Correct answer: D. Canary Deployment.
Canary deployment is ideal for gradually rolling out a new model version to a small, controlled percentage of traffic while simultaneously monitoring its performance. This allows for early detection of issues without impacting the majority of users, aligning perfectly with the requirement to introduce to a small subset and monitor.
Why the other options are wrong
- A. Shadow deployment sends production traffic to both old and new models but only returns the old model's predictions to the user. It's for passive testing, not for actively serving a new model to a subset of users.
- B. A/B testing is for comparing two distinct models or strategies over a longer period, often with different user groups, but doesn't specifically address the gradual rollout to a small percentage of *all* users for a new model version.
- C. Blue/Green deployment switches all traffic at once or in large blocks to a new version, which doesn't fit the 'small subset' requirement.
Canary Deployment (ML)
A deployment strategy where a new model version is rolled out to a small percentage of live traffic first, allowing for real-time monitoring and validation before a full rollout.
- Minimizes risk by exposing new models to a limited audience.
- Enables performance comparison with the old model using live traffic.
- Allows for quick rollback if issues are detected.
Memory trick: Deploy with care, a little at a time, or you'll tear!