Skip to content
Quantoros

Why is a pickled Booster a poor sole long-term archive compared with savemodel?

Classic MLLevel: IntermediateXGBoost · Model IO

Why is a pickled Booster a poor sole long-term archive compared with save_model?

Hints
  • Separate runtime memory from a portable model.
  • Training context still needs its own record.

Answer

A pickle captures implementation-dependent runtime state; a model file stores a stable representation of the trained model. Archive the model plus schema, preprocessing and environment metadata; a model file alone is not the entire reproducible experiment.

Source and licence: XGBoost · Model IO · Apache-2.0 · answer: Quantoros editorial

Related questions