Release 0.4.1 — fix sklearn 1.8 ImportError

CHANGELOG.md

@@ -11,6 +11,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+## [0.4.1] - 2026-05-11
+
+### Fixed
+- `misc/sklearn_compat.py` import failure on `scikit-learn` 1.8+. Upstream renamed `sklearn.utils.validation._is_pandas_df` to `is_pandas_df` (no underscore prefix), which made every `from tabpfn_extensions import ...` immediately `ImportError` on a fresh install resolving sklearn 1.8. Re-vendored `misc/sklearn_compat.py` from the upstream [sklearn-compat](https://github.com/sklearn-compat/sklearn-compat) 0.1.5 release (was 0.1.3, March 2025); the new version has a proper `# Upgrading for scikit-learn 1.8` block plus general cleanups. Only `validate_data` is consumed downstream and its signature hasn't changed. Caught by a TestPyPI smoke install of 0.4.0 before publishing to real PyPI — 0.4.0 was never published.
+
 ## [0.4.0] - 2026-05-11
 
 ### Added

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "tabpfn-extensions"
-version = "0.4.0"
+version = "0.4.1"
 dependencies = [
     "torch>=2.1",
     "pandas>=2.2.2",
@@ -118,6 +118,10 @@ target-version = "py310"
 line-length = 88
 output-format = "full"
 src = ["src", "tests", "examples"]
+# misc/sklearn_compat.py is vendored verbatim from sklearn-compat upstream;
+# excluding it from both lint and format keeps it byte-identical to upstream
+# so future re-vendors are clean diffs.
+extend-exclude = ["src/tabpfn_extensions/misc/sklearn_compat.py"]
 
 [tool.ruff.lint]
 # Allow unused variables when underscore-prefixed.