MLA-C01 Question 366
Select 2You have deployed a machine learning model for real-time inference using Amazon SageMaker. After some time, you notice a significant drop in the model's prediction accuracy. You suspect data drift and want to monitor the incoming data and model predictions to address this issue. Which of the following actions should you take to monitor the model effectively?
- A
Enable Amazon SageMaker Model Monitor to detect data drift in incoming features and labels.
- B
Set up CloudWatch Alarms to monitor the latency of model inference requests.
- C
Use Amazon SageMaker Clarify to monitor and explain feature importance in real time.
- D
Schedule periodic monitoring jobs with SageMaker Model Monitor to evaluate prediction quality over time.
- E
Enable logging of input and output data through Amazon CloudWatch Logs for manual inspection.
Show answer and explanation
Correct answers: A, D
Explanation
To effectively monitor model inference for issues like data drift, SageMaker Model Monitor is the most appropriate tool. It can automatically detect data drift in features and labels with real-time monitoring or periodic monitoring jobs. Other options like latency monitoring or Clarify are designed for different use cases and do not directly address data drift or prediction quality monitoring.
- A. Correct.
Correct: Amazon SageMaker Model Monitor is designed to detect issues like data drift in features and labels, making it a key tool for monitoring model inference.
- B. Incorrect.
Incorrect: Monitoring latency with CloudWatch Alarms is useful for performance issues but does not help in detecting data drift or monitoring prediction quality.
- C. Incorrect.
Incorrect: SageMaker Clarify is used for bias detection and explainability, not for monitoring data drift or model predictions over time.
- D. Correct.
Correct: Scheduling periodic monitoring jobs with SageMaker Model Monitor helps you evaluate prediction quality and detect issues such as data drift or concept drift.
- E. Incorrect.
Incorrect: While logging input and output data can help in manual inspection, it is not an automated or scalable approach for monitoring data drift or prediction quality.