Databricks Machine Learning Professional Question 92
Select 3You are working on a machine learning project where models are deployed using a CI/CD pipeline in Databricks. The team is considering implementing automated testing in the pipeline. Which of the following are valid purposes of automated testing in an ML CI/CD pipeline?
- A
To ensure that the model code produces consistent outputs for the same inputs across different environments.
- B
To prevent models with poor performance from being deployed to production.
- C
To automatically retrain the model when new data is ingested.
- D
To verify that the pipeline components, such as feature engineering and data preprocessing, run as expected.
- E
To reduce the need for manual validation of model predictions in production.
Show answer and explanation
Correct answers: A, B, D
Explanation
Automated testing plays a crucial role in ML CI/CD pipelines by validating the correctness of pipeline components, ensuring consistent model behavior, and enforcing performance thresholds to block suboptimal models from reaching production. However, automated testing does not handle tasks like triggering retraining or completely replacing manual validation, which are distinct aspects of ML workflows.
- A. Correct.
This is correct. Automated testing in an ML CI/CD pipeline helps ensure that the model behaves consistently across different environments, which is crucial for reliable deployments.
- B. Correct.
This is correct. Automated tests in the pipeline can include performance thresholds to prevent poorly performing models from being deployed.
- C. Incorrect.
This is incorrect. Automated testing does not involve triggering model retraining; rather, it focuses on validating the correctness and quality of the pipeline and models.
- D. Correct.
This is correct. Automated testing ensures that all pipeline components, including feature engineering and data preprocessing, function correctly to avoid downstream errors.
- E. Incorrect.
This is incorrect. Automated testing does not eliminate the need for manual validation of model predictions in production, as manual validation can still be critical for certain use cases.