MLA-C01 Question 345
Select 4You are designing a CI/CD pipeline for a machine learning workflow that involves training and deploying a model using AWS services. Which of the following steps should be included to ensure the pipeline adheres to CI/CD principles while addressing the unique needs of ML workflows?
- A
Automate data preprocessing and validation as part of the pipeline.
- B
Integrate model performance monitoring and automated rollback into the deployment phase.
- C
Manually approve every model training job to review hyperparameters before execution.
- D
Use infrastructure as code (IaC) to manage resources such as Amazon SageMaker endpoints and training clusters.
- E
Include automated testing of the model's performance on a holdout dataset before deployment.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
CI/CD principles in ML workflows emphasize automation, consistency, and repeatability. By including steps such as automating data preprocessing, integrating model performance monitoring, managing resources with IaC, and performing automated testing, the pipeline becomes more robust and scalable. However, manual steps like approving every training job contradict the automation goals of CI/CD and should be avoided.
- A. Correct.
Automating data preprocessing and validation ensures consistency and reduces the risk of errors, which aligns with CI/CD principles.
- B. Correct.
Integrating performance monitoring and rollback mechanisms ensures that deployed models can be automatically reverted if performance degrades, which is critical in ML workflows.
- C. Incorrect.
Requiring manual approval for every training job contradicts CI/CD automation principles and would introduce unnecessary delays in the pipeline.
- D. Correct.
Using infrastructure as code (IaC) enables consistent and repeatable provisioning of ML resources, which aligns with CI/CD best practices.
- E. Correct.
Automated testing of model performance ensures that only models meeting predefined thresholds are promoted, aligning with the testing and validation principles of CI/CD.