MLS-C01 Question 373
Select 4You are working on a machine learning pipeline in AWS that processes customer feedback data daily to predict customer sentiment. Over time, the distribution of feedback data has shifted, leading to degraded model performance. You want to implement an automated retraining pipeline that triggers retraining when specific thresholds are met. Which combination of AWS services and features would you use to achieve this?
- A
Amazon SageMaker Model Monitor to detect data drift and trigger retraining workflows
- B
Amazon CloudWatch Alarms to monitor metrics from SageMaker Model Monitor and invoke an AWS Lambda function
- C
AWS Glue to preprocess the data and call SageMaker Model Monitor for drift detection
- D
Amazon SageMaker Pipelines to define a retraining workflow and manage end-to-end automation
- E
AWS Step Functions to manage the orchestration for preprocessing, drift detection, and retraining
Show answer and explanation
Correct answers: A, B, D, E
Explanation
To build an automated retraining pipeline, you need to detect data drift (using SageMaker Model Monitor), monitor drift metrics (using CloudWatch Alarms), and orchestrate the retraining pipeline (using SageMaker Pipelines or AWS Step Functions). AWS Glue is useful for preprocessing but not for drift detection, making it an irrelevant choice for this scenario.
- A. Correct.
Correct: Amazon SageMaker Model Monitor can be used to detect data drift by analyzing input data and model predictions, and it can trigger alerts when drift thresholds are exceeded.
- B. Correct.
Correct: Amazon CloudWatch Alarms can monitor metrics sent by SageMaker Model Monitor, such as drift detection scores, and trigger an AWS Lambda function to initiate retraining.
- C. Incorrect.
Incorrect: While AWS Glue is useful for data preprocessing, it is not responsible for drift detection. SageMaker Model Monitor is specifically designed for detecting data drift.
- D. Correct.
Correct: Amazon SageMaker Pipelines is a fully managed service for defining and automating machine learning workflows, including retraining pipelines.
- E. Correct.
Correct: AWS Step Functions can be used to orchestrate multiple services, such as preprocessing with AWS Glue, drift detection with SageMaker Model Monitor, and retraining with SageMaker.