Skip to content

Model Saving and Loading

MPSBoost uses a versioned native model format.

Supported scope

Currently supported:

  • numeric regression models
  • binary classifier models
  • native CPU multiclass softmax models
  • advanced regression objective metadata

Design constraints

Saved files must contain:

  • model structure
  • version metadata
  • objective metadata
  • feature/bin schema
  • class mapping, for classifiers

Saved files must not contain:

  • training data
  • credentials
  • telemetry
  • device identifiers

Loading performs compatibility validation to prevent reading a model with the wrong estimator or wrong objective.