MLA-C01 Question 379
Select 3You are managing a machine learning model deployed on Amazon SageMaker. Recently, you noticed a decline in the model's prediction accuracy. To proactively monitor the data quality and model performance, which of the following actions should you take?
- A
Enable Amazon SageMaker Model Monitor to track data drift and set up alerts.
- B
Use Amazon CloudWatch to visualize and analyze the model's endpoint invocation metrics.
- C
Retrain the model immediately with new data to improve prediction accuracy.
- D
Perform bias and explainability analysis using Amazon SageMaker Clarify to detect potential issues.
- E
Use a feature store to store and version the input data for the model.
Show answer and explanation
Correct answers: A, B, D
Explanation
Proactively monitoring data quality and model performance requires tools and techniques that provide insights into data drift, bias, and endpoint metrics. Amazon SageMaker Model Monitor helps track and alert on data drift and quality issues, Amazon CloudWatch provides visibility into endpoint metrics, and Amazon SageMaker Clarify helps identify bias and explainability issues. While retraining and using a feature store are important practices in ML workflows, they do not directly contribute to proactive monitoring.
- A. Correct.
Correct. Amazon SageMaker Model Monitor is specifically designed to track data drift, data quality issues, and concept drift. It helps you identify changes in the input data that may affect model performance.
- B. Correct.
Correct. Amazon CloudWatch can be used to monitor endpoint invocation metrics such as latency, throughput, and error rates, which are critical for understanding model performance.
- C. Incorrect.
Incorrect. While retraining the model may be necessary, it is a reactive measure. The question is asking for proactive monitoring techniques to detect issues before retraining is required.
- D. Correct.
Correct. Amazon SageMaker Clarify provides tools for bias detection and explainability, which can help in identifying factors contributing to model performance degradation.
- E. Incorrect.
Incorrect. While a feature store is useful for managing input data and ensuring consistency, it is not a monitoring tool for data quality or model performance.