Release Roadmap

Which features are implemented today and which are planned. Features from the paper not yet in this codebase include bivariate shape functions (\(\mathrm{Shape}^2\mathrm{CART}\)), higher branching factors for bivariate splits, and contour-plot visualization for bivariate splits.

v0.1.0

  • ✅ ShapeCART Classifier & Regressor

  • ✅ Support for higher branching factors (\(\mathrm{SGT}_K\))

  • ✅ Basic plotting via matplotlib

  • ✅ Random forest ensembling for ShapeCART and \(\mathrm{Shape}_K\mathrm{CART}\)

  • ✅ Weighted samples for all SGTs

v0.2.0

  • ✅ Superset branching on categorical features

  • ✅ More plotting options (e.g. exporting to Graphviz)

  • ✅ Feature importances matching scikit-learn’s API for SGTs and \(\mathrm{SGT}_K\)

  • ✅ TAO refinement

  • ✅ Sklearn-style NaN support (replaces current tail-bin placeholder): split search uses finite values only; each candidate is scored with missing sent left vs right—including an explicit missing-vs-non-missing split—with the winning direction stored per node (ties → right).

  • ✅ NaN routing at predict: if training saw missing at that split, follow the stored direction; otherwise route to the majority child.

v0.3.0

  • ⬜ Multioutput support

  • ⬜ \(\mathrm{Shape}^2\mathrm{CART}\)

  • ⬜ \(\mathrm{Shape}^2\mathrm{CART}\) random forest ensembling