Google Professional Machine Learning Engineer Question 412
Select 3Google Cloud PlatformYou are designing a machine learning pipeline on Google Cloud to predict product demand. The model's performance degrades significantly over time due to changes in customer behavior. You want to implement automated model retraining to address this issue. Which of the following considerations are necessary when automating model retraining?
- A
Define a mechanism to monitor model performance and trigger retraining when performance drops below a threshold.
- B
Store a fixed amount of training data for all future retraining jobs to ensure consistency.
- C
Implement versioning for both datasets and models to manage updates and track changes.
- D
Schedule model retraining at fixed intervals regardless of data or performance changes.
- E
Ensure the retraining pipeline includes preprocessing steps to handle data drift or schema changes.
Show answer and explanation
Correct answers: A, C, E
Explanation
Automating model retraining requires careful consideration to ensure the model stays accurate and reliable over time. Key considerations include monitoring model performance to determine when retraining is necessary, versioning datasets and models for reproducibility, and accounting for data drift or schema changes to adapt to evolving data. Fixed schedules or static datasets are less effective as they fail to address dynamic data environments.
- A. Correct.
Correct. Monitoring model performance (e.g., using metrics like accuracy or F1-score) and triggering retraining when performance falls below a threshold ensures the model stays relevant to real-world data.
- B. Incorrect.
Incorrect. Relying on a fixed amount of training data does not account for evolving data distributions. Using dynamic, representative datasets is critical for retraining.
- C. Correct.
Correct. Versioning datasets and models helps maintain reproducibility, track changes, and roll back to previous versions if issues arise.
- D. Incorrect.
Incorrect. Scheduling retraining at fixed intervals without considering data or performance changes can lead to inefficiencies and potentially unnecessary retraining.
- E. Correct.
Correct. Handling data drift or schema changes in the retraining pipeline ensures that the model can adapt to evolving data characteristics.