Google Professional Machine Learning Engineer Question 478
Select 2Google Cloud PlatformYou are a Machine Learning Engineer tasked with deploying a trained classification model to Vertex AI Prediction for production use. The model's predictions will be used in critical decision-making workflows, and stakeholders have requested insights into why the model makes specific predictions. Which steps should you take to enable model explainability on Vertex AI?
- A
Enable the Explainable AI feature when deploying the model to Vertex AI Prediction.
- B
Ensure your model is trained with built-in feature attributions to support explanations.
- C
Use SHAP (SHapley Additive exPlanations) values to explain model predictions after deployment.
- D
Configure explanation parameters during the model deployment process to specify the explanation method and baseline inputs.
- E
Enable AutoML Tables for explainability, as it is the only Vertex AI tool that supports explanations.
Show answer and explanation
Correct answers: A, D
Explanation
To enable model explainability on Vertex AI, you must turn on the Explainable AI feature when deploying your model to Vertex AI Prediction. Additionally, configuring explanation parameters during deployment ensures that the system uses the appropriate method (e.g., integrated gradients, sample-based) and baseline inputs to provide meaningful and accurate explanations for the model's predictions. While other techniques like SHAP values or AutoML Tables support explainability, they are not requirements for using Vertex AI's built-in Explainable AI functionality.
- A. Correct.
Correct: Enabling the Explainable AI feature during deployment is necessary to generate explanations for predictions made by the model in Vertex AI Prediction.
- B. Incorrect.
Incorrect: While feature attributions are used for explainability, you do not need to explicitly train the model with built-in attributions. Explainability is handled during or after deployment.
- C. Incorrect.
Incorrect: SHAP values are a common explainability technique, but their use in Vertex AI is not directly required. Vertex AI has its own built-in explainability mechanisms.
- D. Correct.
Correct: Explanation parameters, such as the explanation method (e.g., integrated gradients) and baseline inputs, must be configured during deployment to tailor the explainability results.
- E. Incorrect.
Incorrect: AutoML Tables supports explainability, but it is not the only Vertex AI tool for this purpose. Custom-trained models on Vertex AI also support Explainable AI.