MLA-C01 Question 259
Select 2You are a Machine Learning Engineer designing a solution to serve a machine learning model in AWS. The model needs to serve predictions in two different ways: low-latency real-time predictions for a web application and periodic batch predictions for generating a daily report. Which combination of services would BEST meet these requirements?
- A
Use Amazon SageMaker Endpoint for real-time predictions and AWS Glue for batch predictions.
- B
Use Amazon SageMaker Endpoint for real-time predictions and Amazon SageMaker Batch Transform for batch predictions.
- C
Use Amazon API Gateway and AWS Lambda for real-time predictions and Amazon EMR for batch predictions.
- D
Use Amazon DynamoDB for real-time predictions and AWS Batch for batch predictions.
- E
Use Amazon SageMaker Multi-Model Endpoint for real-time predictions and Amazon SageMaker Processing for batch predictions.
Show answer and explanation
Correct answers: B, C
Explanation
To serve predictions in both real-time and batch modes, you need to choose AWS services optimized for each case. Amazon SageMaker Endpoint is specifically designed for low-latency, real-time predictions, while SageMaker Batch Transform is built for batch inference tasks. Alternatively, you can use Amazon API Gateway with AWS Lambda for real-time predictions and Amazon EMR for batch processing. These combinations best suit the requirements in the scenario.
- A. Incorrect.
This option uses Amazon SageMaker Endpoint for real-time predictions, which is a suitable choice. However, AWS Glue is primarily used for ETL (Extract, Transform, Load) workflows, not optimized for batch predictions.
- B. Correct.
This option uses Amazon SageMaker Endpoint for real-time predictions and Amazon SageMaker Batch Transform for batch predictions. SageMaker Batch Transform is specifically designed for batch inference, making this a valid choice.
- C. Correct.
This option uses Amazon API Gateway and AWS Lambda for real-time predictions, which is a valid serverless approach for low-latency real-time inference. For batch predictions, Amazon EMR (Elastic MapReduce) can handle large-scale data processing, making this a valid option.
- D. Incorrect.
Amazon DynamoDB is a NoSQL database service and is not designed for serving real-time predictions. AWS Batch, while suitable for batch processing, is not tightly integrated with ML model inference workflows.
- E. Incorrect.
Amazon SageMaker Multi-Model Endpoint is a good option for serving multiple models in real-time, but Amazon SageMaker Processing is designed for data processing and not for batch inference, making this option unsuitable for batch predictions.