modeltrees 0.1.0__tar.gz

modeltrees-0.1.0/.gitignore

@@ -0,0 +1,30 @@
+# Python
+__pycache__/
+*.py[cod]
+.venv/
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+htmlcov/
+dist/
+build/
+*.egg-info/
+
+# R
+.Rhistory
+.RData
+.Rproj.user/
+r/renv/library/
+r/renv/staging/
+r/renv/local/
+
+# Project
+reference/
+notes/
+
+# Editors and OS
+.idea/
+.vscode/
+.DS_Store
+.env

modeltrees-0.1.0/CHANGELOG.md

@@ -0,0 +1,37 @@
+# Changelog
+
+All notable changes are listed here. Versions follow 0.x semantics: names in
+`modeltrees.__all__` and `modeltrees.models` may still change in a minor release
+(docs/api-design.md, section 8.4). A change that alters a grown tree gets an
+entry in docs/decisions.md.
+
+## 0.1.0 (2026-10-22)
+
+First release.
+
+- Model-based recursive partitioning ported from R's partykit 1.3-0: the
+  parameter instability tests with partykit's supLM p-values and chi-squared
+  factor tests, Bonferroni and Sidak adjustment, binary split search for
+  numeric, ordered and nominal partitioning variables, `minsize`, `maxdepth`,
+  `trim`, `restart`, `caseweights` and `numsplit`, and AIC and BIC post-pruning.
+- Entry points `lmtree`, `glmtree` (gaussian, binomial and poisson with
+  canonical links), `negbintree`, `safelogittree`, `multinomtree`, and `mob` for
+  arrays and custom node models.
+- Ports of R's `glm.fit`, `lm.fit`, `lm.wfit` and the LINPACK `dqrdc2` QR, and
+  of MASS's `glm.nb` and `theta.ml`.
+- Negative binomial node model following Rusch (2012), with the log-likelihood
+  (default) or deviance split objective.
+- Safe logit and multinomial logit node models, this project's own provisional
+  designs (docs/extensions-spec.md).
+- Formula interface (`y ~ x1 + x2 | z1 + z2`) on any eager dataframe that
+  narwhals supports, or on a dict of arrays; levels, kinds and missing-data
+  handling as in docs/api-design.md.
+- `cluster` for the instability tests, and cluster-robust leaf covariances
+  (`ModelTree.vcov`, HC0 and HC1 with `cadjust`) checked against
+  `sandwich::vcovCL`.
+- `predict` with types `node`, `link`, `response`, `prob` and `class`;
+  `coef_table`, `summary`, `loglik`, `aic`, `bic` and a printed tree.
+- Parallel split search with `n_jobs`, giving results identical to the serial
+  search under single-threaded BLAS.
+- `modeltrees.reference_path()`, which runs the exact reference implementation
+  that the faster paths are tested against.

modeltrees-0.1.0/CITATION.cff

@@ -0,0 +1,42 @@
+cff-version: 1.2.0
+message: >-
+  If you use modeltrees, please cite the software and the publications
+  whose methods it implements: Zeileis, Hothorn and Hornik (2008) for
+  model-based recursive partitioning, and Rusch (2012) when you use the
+  negative binomial model.
+title: modeltrees
+abstract: >-
+  Model-based recursive partitioning (MOB) in Python, ported from R's
+  partykit, with negative binomial, safe logit and multinomial node models.
+type: software
+version: 0.1.0
+license: GPL-3.0-only
+authors:
+  - alias: BNM
+references:
+  - type: article
+    title: Model-Based Recursive Partitioning
+    authors:
+      - family-names: Zeileis
+        given-names: Achim
+      - family-names: Hothorn
+        given-names: Torsten
+      - family-names: Hornik
+        given-names: Kurt
+    journal: Journal of Computational and Graphical Statistics
+    volume: 17
+    issue: 2
+    start: 492
+    end: 514
+    year: 2008
+    doi: 10.1198/106186008X319331
+  - type: thesis
+    title: Recursive Partitioning of Models of a Generalized Linear Model Type
+    authors:
+      - family-names: Rusch
+        given-names: Thomas
+    thesis-type: Doctoral thesis
+    institution:
+      name: WU Vienna
+    year: 2012
+    doi: 10.57938/f43bea09-fe60-44fb-a995-ba63733c8c69