MLS-C01 Question 7
Single answerYou are building a machine learning system to analyze customer behavior for a retail company. The data consists of structured sales transactions, semi-structured clickstream logs, and unstructured product review images. The team wants to centralize the data in a scalable, cost-effective solution that supports querying, preprocessing, and training ML models. Which solution is best suited for creating the data repository?
- A
Store all data in Amazon S3, use S3 Select for querying structured data, and integrate with AWS Glue for preprocessing.
- B
Store structured data in Amazon RDS, semi-structured data in DynamoDB, and unstructured data in Amazon S3, then preprocess them separately.
- C
Store all data in an Amazon Redshift data warehouse and use Amazon Redshift ML for preprocessing and training.
- D
Store all data in Amazon S3, catalog the data with AWS Glue Data Catalog, and preprocess using Amazon SageMaker Processing.
Show answer and explanation
Correct answer: D
Explanation
The correct solution is to store all data in Amazon S3 because it is scalable, cost-effective, and supports all data types (structured, semi-structured, and unstructured). Cataloging with AWS Glue Data Catalog allows for efficient querying and data discovery, while Amazon SageMaker Processing offers a unified and managed approach to preprocessing the data before training ML models. This approach aligns with best practices for creating data repositories for machine learning in AWS.
- A. Incorrect.
This is a partially correct approach as Amazon S3 and S3 Select are cost-efficient for querying structured data, but it does not address semi-structured or unstructured data effectively. Additionally, it lacks a cohesive strategy for preprocessing and training.
- B. Incorrect.
This approach involves using multiple services for different data types, which increases complexity and may incur higher costs. It also lacks a unified data catalog for efficient querying and preprocessing.
- C. Incorrect.
Amazon Redshift is primarily designed for structured and semi-structured data, and while it supports Amazon Redshift ML, it is not ideal for handling unstructured data like images. This solution is not optimal for centralizing all data types.
- D. Correct.
Amazon S3 is a cost-effective and scalable storage solution for all data types. By cataloging the data with AWS Glue Data Catalog, you can enable efficient querying and data discovery. Amazon SageMaker Processing provides a managed solution for preprocessing that integrates well with the ML workflow.