MLS-C01 Question 376
Select 3A company has deployed a machine learning model for predicting customer churn using Amazon SageMaker. The model performance has started to degrade due to changes in customer behavior. The data science team wants to set up an automated pipeline to retrain the model using new data collected daily. Which of the following steps are necessary to create this retraining pipeline effectively?
- A
Set up an Amazon SageMaker Processing job to preprocess the new data before training.
- B
Use Amazon CloudWatch Events to trigger the retraining pipeline based on a predefined schedule.
- C
Configure Amazon SageMaker Model Monitor to automatically update the model without retraining.
- D
Store the new training data in an Amazon S3 bucket accessible by the SageMaker pipeline.
- E
Replace the existing endpoint with a new one manually after training the updated model.
Show answer and explanation
Correct answers: A, B, D
Explanation
To set up an effective retraining pipeline, you need to preprocess the new data (using SageMaker Processing), store the data in an accessible location (e.g., S3), and automate the pipeline's execution (e.g., using CloudWatch Events). SageMaker Pipelines can handle model deployment after retraining, so manual intervention to replace endpoints is unnecessary. SageMaker Model Monitor, while useful for identifying data drift, does not retrain models automatically.
- A. Correct.
Correct: Preprocessing the data is an essential step in any machine learning pipeline. Amazon SageMaker Processing jobs allow you to handle tasks like data cleaning, feature engineering, and transformations before model training.
- B. Correct.
Correct: Amazon CloudWatch Events (now called EventBridge) can be used to schedule and trigger the retraining pipeline automatically at regular intervals or based on specific events.
- C. Incorrect.
Incorrect: Amazon SageMaker Model Monitor is used to monitor data and model quality but does not retrain or update models automatically.
- D. Correct.
Correct: The retraining pipeline requires access to the new training data, which should be stored in an accessible location like an Amazon S3 bucket.
- E. Incorrect.
Incorrect: Replacing the endpoint manually is not required if you use Amazon SageMaker Pipelines. The pipeline can automate model deployment after retraining.