Linearmodels

This is a compatibility release.

* Compatibility with NumPy 2
* Compatibility with recent pandas releases

* Bumped the minimum formulaic to 0.6.5.
* Released wheels for Python 3.12.

* Fixed a bug that affected dropping absorbed regressors when using weights (546).
* Improved the licensing information (545).
* Improved compatibility with recent pandas (548).

This is a bug-fix release.

* Fixes a bug that can be encountered when collection observation data in panel models when some entities or time periods are missing.

This major release contains breaking changes when using formulas and increases the requirements for running linear models.

- The variable order is preserved when creating a model using `from_formula`. Previously variables were sorted irrespective of the order they appeared in the formula.
- Increased minimums:

- Python: 3.9
- formulaic: 0.6.1
- NumPy: 1.19.0
- SciPy: 1.5.0
- pandas: 1.1.0
- statsmodels: 0.12.0

* Switched variable ordering by default. Importing linearmodels.__future__.ordering is a no-op, and has no effect.
* Removed dependence on property-cached in favor of [:meth:`functools.cached_property`](https://github.com/bashtage/linearmodels/blob/main/doc/source/changes/5.0.rst#id1).

This release adds support for formulaic 0.6.0. This version of formulaic makes a significant change to how variables from formulas are translated into DataFrames for use in models. The existing behavior sorts variables. The new behavior respects the order of the variables as they appear. The new behavior will become the default in linearmodels starting in release 5.

To use the new behavior now, add the import

python
from linearmodels.__future__ import ordering


to the top of your file.