MLA-C01 Question 362
Select 3An e-commerce company has deployed a machine learning model using Amazon SageMaker to recommend products to customers based on their browsing and purchase history. Over time, the Data Science team notices a decline in the model's recommendation performance due to changes in customer behavior. Which action should the team take to monitor and address this issue effectively?
- A
Enable Amazon SageMaker Model Monitor to track data drift and model performance metrics.
- B
Set up an AWS Lambda function to retrain the model in real-time whenever it detects performance degradation.
- C
Use Amazon CloudWatch to collect and analyze model logs for identifying anomalies in predictions.
- D
Schedule periodic retraining of the model using updated data in Amazon SageMaker Pipelines.
- E
Implement endpoint invocation logging in SageMaker to capture input data and prediction outputs for debugging.
Show answer and explanation
Correct answers: A, D, E
Explanation
To effectively monitor and address the declining performance of a machine learning model, it is critical to track data drift and model performance using tools like SageMaker Model Monitor. Periodic retraining ensures the model adapts to new data patterns, while invocation logging helps diagnose issues by capturing inputs and outputs. Although CloudWatch is useful for general logging, it does not specifically address model performance degradation, and real-time retraining with Lambda is not a practical solution.
- A. Correct.
Correct: Amazon SageMaker Model Monitor can be used to track data drift, feature drift, and other metrics to ensure the model's performance is not degrading due to changes in input data.
- B. Incorrect.
Incorrect: Real-time retraining using AWS Lambda is not a recommended or practical approach for retraining ML models, as it may lead to resource inefficiency and overfitting.
- C. Incorrect.
Incorrect: While Amazon CloudWatch can be used for logging and monitoring, it is not specifically designed for detecting or addressing model performance degradation.
- D. Correct.
Correct: Periodic retraining using updated data through Amazon SageMaker Pipelines ensures the model remains up-to-date with changing data patterns.
- E. Correct.
Correct: Enabling endpoint invocation logging in SageMaker provides insights into the model's inputs and outputs, helping to diagnose performance issues and data drift.