MLS-C01 Question 173
Select 3A retail company uses an Amazon SageMaker endpoint to serve a machine learning model for demand forecasting. The data distribution has recently shifted due to seasonal changes, resulting in decreased model accuracy. What steps should you take to update and retrain the model using Amazon SageMaker, while minimizing downtime?
- A
Collect new data reflecting recent seasonal trends and use it to retrain the model.
- B
Deploy a new version of the model to the same endpoint using Amazon SageMaker's multi-model endpoint functionality.
- C
Use Amazon SageMaker Model Monitor to detect data drift and trigger a retraining workflow.
- D
Retrain the model using the original dataset to ensure consistency in predictions.
- E
Perform A/B testing by deploying the updated model to a shadow endpoint and monitoring performance before switching traffic.
Show answer and explanation
Correct answers: A, C, E
Explanation
To address a shift in data distribution and retrain a model, you need to collect new data that reflects the recent changes, ensure data drift is detected with tools like SageMaker Model Monitor, and safely validate the updated model using A/B testing before deploying it fully to production. These steps minimize downtime and ensure reliability in your updated model's predictions.
- A. Correct.
Correct: Collecting new data reflecting seasonal trends is necessary to ensure the model is updated with the latest patterns in the data.
- B. Incorrect.
Incorrect: Multi-model endpoints are designed for hosting multiple models behind a single endpoint, not for updating an existing model with new training data.
- C. Correct.
Correct: Amazon SageMaker Model Monitor can be used to detect data drift, which is critical for identifying when retraining is needed. It can also trigger workflows for retraining.
- D. Incorrect.
Incorrect: Retraining the model using only the original dataset will not address the shift in data distribution caused by seasonal changes.
- E. Correct.
Correct: A/B testing with a shadow endpoint allows for safe validation of the updated model without impacting current users. This minimizes risk before fully shifting traffic to the new model.