Google Professional Machine Learning Engineer Question 89
Select 4Google Cloud PlatformYou have deployed a machine learning model on Google Cloud AI Platform (Vertex AI). However, during testing, you observe that the model's predictions are significantly less accurate than expected. You want to debug and improve the model. Which of the following actions should you take to identify and fix the issue?
- A
Analyze the model's feature importance to understand which features contribute most to predictions.
- B
Enable Vertex AI Prediction request and response logging to inspect inputs and outputs.
- C
Directly adjust the model's weights using Vertex AI Console to improve accuracy.
- D
Perform data validation to check for issues such as missing values or data drift.
- E
Use Explainable AI on Vertex AI to interpret the model's predictions and identify potential biases.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
Debugging a trained model involves identifying issues with the data, features, or modeling process, and addressing them systematically. Analyzing feature importance, inspecting prediction logs, validating data, and using Explainable AI are all critical tools for diagnosing and resolving performance issues. Manually adjusting model weights is not a practical or standard debugging practice.
- A. Correct.
Analyzing feature importance is a crucial step to debug model performance. It helps you understand if the model relies on the right features or if irrelevant features are dominating predictions.
- B. Correct.
Enabling request and response logging allows you to inspect the actual data being passed to the model at inference time. This can help identify if the inputs or outputs are misaligned with expectations.
- C. Incorrect.
Directly adjusting model weights manually is not a standard or recommended debugging practice in Vertex AI or machine learning in general. Model retraining with updated data is the correct approach.
- D. Correct.
Data validation is essential to ensure the input data is clean, consistent, and free of issues like missing values or data drift, which can significantly degrade model performance.
- E. Correct.
Explainable AI provides insights into the model's decision-making process. This can help identify biases or unexpected behavior in the model's predictions.