Changelog
This log follows the package versions declared in pyproject.toml.
v0.4.3 — 2026-06-29
- Published the canonical supervised-region API from the package root so fresh and upgraded Colab runtimes can import all three region clusterers.
- Made the packaged Colab notebook install the exact release, verify its exports, and run independently of repository-local result files.
- Added a self-contained Diabetes continuous-target example with regional η², coverage, unmatched rate, dispersion reduction, and separate forest R²/RMSE diagnostics.
- Normalized low-level multiclass parameters to
leaf_percentile,low_leaf_fraction,max_regions_per_class, andrule_score; legacy spellings remain available withFutureWarning. - Confirmed that deprecated classes, methods, and attributes remain migration aliases throughout 0.4.x and are scheduled for removal in 0.5.0.
- Added wheel-level import, stale-install upgrade, and isolated notebook execution regression coverage.
Version note: 0.4.2 was not published; 0.4.3 is the first public release after the internal 0.4.1 API transition.
v0.4.1 — 2026-06-28
- Redefined the canonical estimators as supervised region clusterers:
InsideForestRegionClusterer,InsideForestClassRegionClusterer, andInsideForestContinuousRegionClusterer. - The forest now generates candidate leaves only.
predictreturns cluster IDs, unmatched observations receive-1, and the canonical API has no class/value prediction fallback. - Added canonical region tables, assignment details, transformations, quality reports, persistence, class-aware diagnostics, continuous-target statistics, AMI scoring for categorical targets, and η² scoring for continuous targets.
- Reserved
labels_for training cluster assignments and moved per-region measurements toregion_metrics_. - Made branch aggregation opt-in and initialized the canonical estimators with
branch_aggregation="none". - Deprecated
InsideForestClassifier,InsideForestMulticlassClassifier, andInsideForestRegressoras migration aliases. - Updated the bilingual documentation, Colab notebook, benchmarks, package metadata, migration notes, and acceptance tests for the regional contract.
v0.4.0 — 2026-06-06
- Previous public package release, based on the legacy classifier/regressor interfaces.