Google Professional Machine Learning Engineer Question 489
Select 2Google Cloud PlatformYou are deploying a real-time fraud detection model on Google Cloud using Vertex AI. The model's performance may degrade over time due to changes in user behavior. How can you ensure continuous monitoring and evaluation of the model's effectiveness while keeping resource usage efficient?
- A
Configure Vertex AI Model Monitoring to track feature skew and drift in input data.
- B
Set up a pipeline to automatically retrain the model whenever drift is detected.
- C
Manually evaluate the model's predictions on a monthly basis to check for accuracy degradation.
- D
Enable Vertex AI Model Monitoring to track prediction drift and alert you when it exceeds a specified threshold.
- E
Deploy a secondary model to cross-check the predictions of the primary model in real-time.
Show answer and explanation
Correct answers: A, D
Explanation
Vertex AI Model Monitoring provides a robust way to track data and prediction drift, ensuring your model's performance is evaluated continuously. By tracking feature skew, input data drift, and prediction drift, you can identify issues early and take corrective actions. Setting alerts for drift thresholds ensures timely intervention without unnecessary resource consumption.
- A. Correct.
Correct. Vertex AI Model Monitoring can track feature skew and drift in input data, which helps identify changes in the data distribution that may degrade model performance.
- B. Incorrect.
Incorrect. While retraining is important, automatically retraining the model without validation or understanding the cause of drift can lead to unintended consequences. Monitoring and analysis come first.
- C. Incorrect.
Incorrect. Manual evaluation is time-consuming and inefficient for real-time systems. Automated monitoring solutions are better suited for this scenario.
- D. Correct.
Correct. Enabling Vertex AI Model Monitoring to track prediction drift and set alerts for thresholds ensures you are notified of performance degradation promptly.
- E. Incorrect.
Incorrect. Deploying a secondary model for real-time cross-checking is not a resource-efficient approach and is not a standard practice for continuous monitoring.