AIF-C01 Question 205
Select 3A company is building a machine learning model using Amazon SageMaker. The data science team wants to ensure that they can effectively track the origins of their datasets, transformations applied, and the model's training history. Which combination of tools or practices should they use to achieve this?
- A
Use SageMaker Model Cards to document the model's training history and data sources.
- B
Maintain a manual spreadsheet to document data transformations and sources.
- C
Implement a data catalog using AWS Glue to track data origins and metadata.
- D
Leverage SageMaker Lineage Tracking to monitor data and model dependencies.
- E
Rely solely on the training script comments to document data origins.
Show answer and explanation
Correct answers: A, C, D
Explanation
To effectively track data origins and maintain transparency in machine learning workflows, it is essential to use integrated tools like SageMaker Model Cards, AWS Glue Data Catalog, and SageMaker Lineage Tracking. These tools provide structured and scalable solutions for documenting data lineage, transformations, and model histories. Manual methods or relying solely on training script comments are not sufficient for ensuring traceability and scalability.
- A. Correct.
SageMaker Model Cards are used to document details about a model, such as its purpose, training datasets, evaluation metrics, and lineage. This is a key tool for tracking the model's history.
- B. Incorrect.
While maintaining a manual spreadsheet can help with documentation, it is prone to errors, lacks automation, and does not scale well for tracking complex data and model dependencies.
- C. Correct.
AWS Glue Data Catalog allows you to organize and track metadata for datasets, making it easier to identify the origins and structure of data used in machine learning pipelines.
- D. Correct.
SageMaker Lineage Tracking is explicitly designed to track dependencies between datasets, transformations, and models, ensuring transparency and traceability.
- E. Incorrect.
Training script comments are insufficient for comprehensive tracking of data origins and transformations as they are not structured or integrated with tools that manage lineage and metadata.