問題文
A model needs a preprocessing step that maps free-text product names to canonical categories using a lookup file, plus a post-processing step that rounds the output to two decimal places. The team wants the served endpoint to apply both. How should the model be packaged?
選択肢
- As the plain estimator, with the mapping expressed as a Unity Catalog function that the endpoint calls before inference, one call per name.
- As a custom Python model class whose prediction method performs the mapping, calls the wrapped estimator, and applies the rounding, logged together with the lookup file as an artifact.
- As two separate endpoints chained by the application, one for preprocessing and one for the estimator, called in order.
- As the plain estimator, with the mapping and rounding implemented in the calling application so that the model stays simple, with the lookup file held by it.