You have trained an XGBoost model that you plan to deploy on Vertex AI for online prediction. You are now uploading your model to Vertex AI Model Registry, and you need to configure the explanation…

Question

Options

• A
  1. Specify sampled Shapley as the explanation method with a path count of 5.
• B
  1. Specify Integrated Gradients as the explanation method with a path count of 5.
• C
  1. Specify sampled Shapley as the explanation method with a path count of 50.
• D
  1. Specify Integrated Gradients as the explanation method with a path count of 50.

• A
  72% (33)
• B
  4% (2)
• C
  15% (7)
• D
  9% (4)

Explanation

For minimal latency online predictions with explanations, two choices matter: (1) Sampled Shapley is faster than Integrated Gradients because it approximates SHAP values with random sampling rather than computing integrals along feature paths - better for low-latency requirements. (2) A low path count (5) reduces computation per request at the cost of some approximation accuracy; for online serving where speed is critical, this tradeoff is correct. Integrated Gradients (Options B, D) is more accurate but slower, and is better suited for tabular models only when latency is not a constraint. A path count of 50 (Options C, D) increases computation time. After configuring explanations, Vertex AI Model Monitoring can track feature attribution drift and send alerts when attributions change meaningfully over time.

No community discussion yet for this question.