Skranger

~~~~~~~~~~~~~~~~~~

* Replace poetry with poetry-core as a build dependency

~~~~~~~~~~~~~~~~~~~~~

* Drop support for Python 3.7
* Add support for Python 3.10
* Add support for Windows
* Update scikit-learn constraint to 1.0

~~~~~~~~~~~~~~~~~~

* Drop support for Python 3.6
* Fix quantile regression predictions for single record
* Fix Tree.value for classifiers
* Fix Tree.feature to use proper value for leaf nodes
* Allow ``skranger`` predictors to work with shap using ``skranger.utils.shap.shap_patch`` context manager
* Fix package includes to prevent installing extra files to ``site-packages``

~~~~~~~~~~~~~~~~~~

* Use oldest supported numpy for builds

~~~~~~~~~~~~~~~~~~

* Remove numpy from dependency spec; numpy is already a requirement of scikit-learn
* Change tree detail training code to be optional in ensembles due to expensive operations
* Change quantile regression to use ``np.quantile`` in lieu of `np.nanquantile`` for faster predictions
* Fix bug in tree classmethods when setting ``sample_fraction``
* Added more documentation around tree detail calculations

~~~~~~~~~~~~~~~~~~

* Move ``split_select_weights``, ``always_split_features``, ``categorical_features`` params from init to fit methods.
* Sample weight checking is now a base class method.
* Remove sparse matrix args from bindings.
* Fix a bug with the output of ``predict_quantiles`` not being oriented properly for multiple quantiles
* Regression's ``predict_quantiles`` now requires a passed list of quantiles and the default is removed
* Regression's ``predict`` now takes an optional list of quantiles
* Remove ``snp_data`` and ``order_snps`` from bindings
* Moves ``class_weights`` to fit in classifier, and changes the arg type to a dictionary.
* Add ``RangerTreeClassifier``, ``RangerTreeRegressor``, and ``RangerTreeSurvival`` decision tree estimators which inherit between ``RangerMixin`` and ``BaseRangerTree``. Also provide a ``BaseRangerForest`` class for ensemble estimators.
* Add a low level ``Tree`` class which implements most of the ``sklearn.tree._tree.Tree`` interface.
* Fix incorrect documentation for ``num_random_splits``.