AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsMedium
A data science team is developing a critical machine learning model for real-time anomaly detection. They want to introduce a new version of the model to production without directly impacting live users and gather performance metrics against real production traffic. Only after thorough evaluation with live data will they consider switching over. Which deployment strategy should they use?
- AA/B Testing
- BBlue/Green Deployment
- CShadow Deployment
- DCanary Deployment
Show answer & explanationAnswer & explanation
Correct answer: C. Shadow Deployment
Shadow deployment allows a new model version to process a copy of live production traffic in parallel with the current production model, without its predictions impacting the actual live responses. This enables thorough evaluation of the new model's performance on real data before it's ever exposed to users, perfectly matching the team's requirement to not impact live users while gathering metrics.
Why the other options are wrong
- A. A/B testing is used to compare two models by routing different user groups to each, meaning both models are live and impacting users.
- B. Blue/Green deployment switches all traffic at once, which would impact live users.
- D. Canary deployment routes a small percentage of live traffic to the new model, meaning its predictions would impact those users.
Shadow Deployment
A deployment strategy where a new model version runs in parallel with the currently deployed model, processing a copy of live production traffic, but its predictions are not used by the application.
- New model's predictions do not impact live users.
- Allows safe evaluation of new model performance with real-world data.
- Useful for testing stability, latency, and accuracy under production load.
Memory trick: Shadows watch, never act.