MLS-C01 Question 358
Single answerA company has developed a machine learning model to predict customer churn and needs to deploy it to production for real-time inference. The model is trained on SageMaker and needs to handle unpredictable traffic patterns while ensuring high availability. Additionally, the team wants to monitor the model's performance and log inference requests and responses for future analysis. What is the best deployment solution using AWS services?
- A
Deploy the model using Amazon SageMaker Hosting Services with auto-scaling enabled and configure Amazon CloudWatch for monitoring.
- B
Deploy the model on an EC2 instance with an Elastic Load Balancer and manually scale the instances as needed.
- C
Use AWS Lambda to deploy the model and integrate it with Amazon S3 for logging inference requests and responses.
- D
Deploy the model using Amazon SageMaker Batch Transform for handling high traffic and use AWS Config for monitoring.
Show answer and explanation
Correct answer: A
Explanation
Amazon SageMaker Hosting Services is specifically designed for deploying ML models in production environments. It supports real-time inference, auto-scaling for fluctuating traffic, and integrates with Amazon CloudWatch for monitoring and logging. This makes it the most effective solution for the given scenario, ensuring high availability and operational ease.
- A. Correct.
This is the correct option. SageMaker Hosting Services supports deploying models for real-time inference with auto-scaling to handle varying traffic patterns. It can also be integrated with Amazon CloudWatch for monitoring, making it the best choice for this scenario.
- B. Incorrect.
While deploying on EC2 with an Elastic Load Balancer is possible, it requires significant manual effort to handle scaling and maintenance. It is not the ideal solution for an ML model requiring real-time inference with high availability.
- C. Incorrect.
AWS Lambda is not suitable for deploying large ML models requiring persistent runtime environments for inference. Additionally, Lambda has limitations on request payload size and execution time, making it less ideal for this use case.
- D. Incorrect.
Amazon SageMaker Batch Transform is designed for batch processing rather than real-time inference. It is not appropriate for handling unpredictable traffic patterns or real-time requests, and AWS Config is not used for model performance monitoring.