Question 270
A model deployment uses custom preprocessing before batch inference and the deployment skips that stage. The team can modify the relevant resource but must not broaden unrelated permissions or access. The sign-off condition is that batch predictions match a fixed validated fixture after preprocessing. What is the most appropriate corrective action?
Explanation
Correct Answer
The serving prediction function needs the same feature contract established during training. On retest, batch predictions match a fixed validated fixture after preprocessing.
Incorrect Answers
A batch job's runtime is independent of the development notebook environment. The serving prediction function needs the same feature contract established during training.
Batch endpoint results can contain sensitive data and require separate output storage controls.
Long-running bulk scoring is a batch concern rather than a synchronous request-response service.