Why should a native-categorical XGBoost model be round-trip tested using JSON or UBJSON?
Classic MLLevel: IntermediateXGBoost · Categorical data
Why should a native-categorical XGBoost model be round-trip tested using JSON or UBJSON?
Hints
- A category split needs more than a numeric threshold.
- Serialization must retain that metadata.
Answer
The supported formats preserve categorical metadata needed to interpret splits. Compare predictions after loading on known and boundary category cases; a serialized tree without its category semantics is not a reliable deployment artifact.
Source and licence: XGBoost · Categorical data · Apache-2.0 · answer: Quantoros editorial