v3.3.1

This is a bugfix release.

Fixes

• Make skl2onnx optional (#21)
• Expose minimal required versions for ir and opset (#22, #23)
• Fix multiclass classification with interactions (#24)

v3.3.0

This is an improvement release.

Improvements

• Initial support for model serialization within a scikit-learn pipeline (#9)
• Added support for loading an existing ONNX model. This allows for editing an existing model.

v3.2.0

This is a bugfix and improvement release.

Improvements

• Add numpy 2.0 compatibility (#17)

Fixes

• Fixed a regression introduced by #12 that mutates the original ebm model (#16)

v3.1.3

This is a bugfix release.

Fixes

• The conversion fails if a boolean feature has only one value (#11)

v3.1.2

This is a bugfix release.

Fixes

• Boolean data column leads to wrong predictions (#11)

v3.1.1

This is a bugfix release.

Fixes

• The output value of a classification is an index instead of the class (#6)

v3.1.0

This is an improvement release with breaking changes.

• The predict_proba parameter now creates a dedicated output, in addition to the prediction output.
• The names of the outputs (prediction, probabilities, and explanation) are now configurable.

v2.0.0

This is an improvements release, with breaking changes:

This version depends on at least Interpret v0.3.0 where the internal representation of the EBM models changed.

v1.3.0

This is an improvements release.

Improvements

• Fix the name of the scores and predict_proba outputs (#3). They are now named "scores_0" and "predict_proba_0"
• Add support for categorical features of any type (#1)

v1.2.0

This is an improvement release.

Improvements

• Add a pandas dtype helper to create the dtype parameter automatically (#2).
• Add support for boolean continuous features.
• Print an explicit error on unsupported categorical feature types.