Google Professional Machine Learning Engineer Question 479
Select 2Google Cloud PlatformYou are deploying a machine learning model on Vertex AI Prediction and want to ensure compliance with regulatory requirements by providing explanations for predictions made by the model. Which steps should you take to enable model explainability on Vertex AI?
- A
Use a model format that supports explainable AI, such as TensorFlow SavedModel or XGBoost, and deploy it to a Vertex AI endpoint with explanation metadata.
- B
Enable the Explainability feature during model deployment and provide an explanation configuration that specifies the input fields to explain.
- C
Deploy the model with a custom container and implement a custom explanation API to return feature attributions.
- D
Use SHAP or Integrated Gradients in your training pipeline and ensure the attributions are stored alongside predictions in the model database.
- E
Configure Vertex AI Prediction to automatically enable explainability while deploying models without additional configuration.
Show answer and explanation
Correct answers: A, B
Explanation
To ensure model explainability on Vertex AI, you must use supported model formats (e.g., TensorFlow SavedModel, XGBoost) and enable the Explainability feature during deployment. This includes specifying explanation metadata and input fields for generating feature attributions. Custom implementation or offline tools like SHAP are not required for Vertex AI Prediction explainability.
- A. Correct.
Correct: Vertex AI supports explainability for models in formats like TensorFlow SavedModel and XGBoost, provided explanation metadata is included during deployment.
- B. Correct.
Correct: Enabling the Explainability feature and providing an explanation configuration is a required step to generate feature attributions in Vertex AI.
- C. Incorrect.
Incorrect: While you can deploy a model with a custom container, implementing a custom explanation API is not required or aligned with Vertex AI's built-in explainability features.
- D. Incorrect.
Incorrect: Tools like SHAP and Integrated Gradients are useful during training or offline analysis, but they do not directly integrate with Vertex AI Prediction for generating online explanations.
- E. Incorrect.
Incorrect: Automatic explainability is not enabled by default in Vertex AI; it requires explicit configuration during deployment.