Google Professional Machine Learning Engineer Question 482
Select 3Google Cloud PlatformYou are designing a machine learning model deployment on Vertex AI Prediction and need to ensure that stakeholders can understand the model's decisions. As part of the process, you want to utilize Vertex AI's model explainability features. Which of the following steps are required to enable model explainability for your deployed model?
- A
Enable model explanation during the deployment process by specifying an explanation method, such as SHAP or Integrated Gradients.
- B
Use a pre-trained explainability model provided by Vertex AI to generate explanations for your predictions automatically.
- C
Ensure that your input data includes feature metadata and that you configure the deployed model with feature attributions.
- D
Deploy the model on Vertex AI with an endpoint that has explanation metadata defined in its configuration.
- E
Train the model exclusively on tabular data, as Vertex AI explainability features are only supported for tabular models.
Show answer and explanation
Correct answers: A, C, D
Explanation
To enable model explainability on Vertex AI Prediction, you need to specify an explanation method during deployment, configure the model with feature metadata for feature attributions, and ensure that explanation metadata is included in your endpoint configuration. These steps allow Vertex AI to generate detailed explanations of model predictions, which are crucial for building trust with stakeholders.
- A. Correct.
Correct: Vertex AI supports explanation methods like SHAP and Integrated Gradients, which need to be explicitly enabled during the deployment process.
- B. Incorrect.
Incorrect: Vertex AI does not provide pre-trained explainability models. Instead, it uses methods like feature attributions or SHAP to explain predictions.
- C. Correct.
Correct: Model explainability on Vertex AI requires configuring the model with feature metadata to calculate attributions for input features.
- D. Correct.
Correct: Explanation metadata must be defined in the endpoint configuration when deploying the model on Vertex AI.
- E. Incorrect.
Incorrect: Vertex AI explainability features are supported for various data types, including tabular, image, and text data, not just tabular data.