Google Professional Machine Learning Engineer Question 490
Select 3Google Cloud PlatformYou have deployed a machine learning model on Vertex AI to predict loan defaults. The model is used in a production environment, and you need to ensure its performance is monitored over time to detect issues such as data drift and feature skew. Which steps should you take to implement continuous evaluation using Vertex AI Model Monitoring?
- A
Enable model monitoring in Vertex AI and configure alerting thresholds for data drift and feature skew.
- B
Use Vertex AI Pipelines to retrain the model on a fixed schedule without monitoring data drift.
- C
Define a baseline dataset and set feature attribution thresholds for monitoring.
- D
Configure a BigQuery table to store prediction requests and actual labels for evaluation.
- E
Set up Cloud Logging to monitor prediction latency and system performance metrics.
Show answer and explanation
Correct answers: A, C, D
Explanation
To implement continuous evaluation using Vertex AI Model Monitoring, it's important to enable model monitoring, define a baseline dataset, and set monitoring thresholds for detecting issues like data drift and feature skew. Storing prediction requests and actual labels in a BigQuery table provides the data necessary for detailed evaluation. While system performance monitoring is useful, it doesn't replace the need for specific model monitoring configurations.
- A. Correct.
Correct. Enabling model monitoring in Vertex AI and configuring thresholds for data drift and feature skew is an essential step in continuous evaluation.
- B. Incorrect.
Incorrect. Retraining the model on a fixed schedule without monitoring doesn't address the need for continuous evaluation of data drift or feature skew.
- C. Correct.
Correct. Defining a baseline dataset and setting feature attribution thresholds is critical for monitoring changes in data patterns and ensuring model reliability.
- D. Correct.
Correct. Configuring a BigQuery table to store prediction requests and actual labels allows for detailed analysis and comparison for continuous evaluation.
- E. Incorrect.
Incorrect. While Cloud Logging is useful for monitoring system performance, it doesn't directly assist in evaluating model performance, data drift, or feature skew.