.. include:: _contributors.rst

.. currentmodule:: sklearn

.. _release_notes_1_9:

===========
Version 1.9
===========

..
  -- UNCOMMENT WHEN 1.9.0 IS RELEASED --
  For a short description of the main highlights of the release, please refer to
  :ref:`sphx_glr_auto_examples_release_highlights_plot_release_highlights_1_9_0.py`.


..
  DELETE WHEN 1.9.0 IS RELEASED
  Since October 2024, DO NOT add your changelog entry in this file.
..
  Instead, create a file named `<PR_NUMBER>.<TYPE>.rst` in the relevant sub-folder in
  `doc/whats_new/upcoming_changes/`. For full details, see:
  https://github.com/scikit-learn/scikit-learn/blob/main/doc/whats_new/upcoming_changes/README.md

.. include:: changelog_legend.inc

.. towncrier release notes start

.. _changes_1_9_dev0:

Version 1.9.dev0
================

**February 2026**

Changed models
--------------

- |Enhancement| The :meth:`transform` method of :class:`preprocessing.PowerTransformer` with
  `method="yeo-johnson"` now uses the numerical more stable function
  `scipy.stats.yeojohnson` instead of an own implementation. The results may deviate in
  numerical edge cases or within the precision of floating-point arithmetic.
  By :user:`Christian Lorentzen <lorentzenchr>`. :pr:`33272`

Changes impacting many modules
------------------------------

- |Enhancement| :class:`pipeline.Pipeline`, :class:`pipeline.FeatureUnion` and
  :class:`compose.ColumnTransformer` now raise a clearer
  error message when an estimator class is passed instead of an instance.
  By :user:`Anne Beyer <AnneBeyer>` :pr:`32888`

- |Fix| Raise ValueError when `sample_weight` contains only zero values to prevent
  meaningless input data during fitting. This change applies to all estimators that
  support the parameter `sample_weight`. This change also affects metrics that validate
  sample weights.
  By :user:`Lucy Liu <lucyleeow>` and :user:`John Hendricks <j-hendricks>`. :pr:`32212`

- |Fix| Some parameter descriptions in the HTML representation of estimators
  were not properly escaped, which could lead to malformed HTML if the
  description contains characters like `<` or `>`.
  By :user:`Olivier Grisel <ogrisel>`. :pr:`32942`

Support for Array API
---------------------

Additional estimators and functions have been updated to include support for all
`Array API <https://data-apis.org/array-api/latest/>`_ compliant inputs.

See :ref:`array_api` for more details.

- |Feature| :func:`sklearn.metrics.d2_absolute_error_score` and
  :func:`sklearn.metrics.d2_pinball_score` now support array API compatible inputs.
  By :user:`Virgil Chan <virchan>`. :pr:`31671`

- |Feature| :class:`linear_model.LogisticRegression` now supports array API compatible inputs
  with `solver="lbfgs"`.
  By :user:`Omar Salman <OmarManzoor>` and :user:`Olivier Grisel <ogrisel>`. :pr:`32644`

- |Feature| :func:`sklearn.metrics.ranking.average_precision_score` now supports Array API
  compliant inputs.
  By :user:`Stefanie Senger <StefanieSenger>`. :pr:`32909`

- |Feature| :func:`sklearn.metrics.pairwise.paired_manhattan_distances` now supports array API
  compatible inputs. By :user:`Bharat Raghunathan <bharatr21>`. :pr:`32979`

- |Feature| :func:`sklearn.metrics.pairwise.pairwise_distances_argmin` now supports array API
  compatible inputs. By :user:`Bharat Raghunathan <bharatr21>`. :pr:`32985`

- |Feature| :class:`pipeline.FeatureUnion` now supports Array API compliant inputs when all
  its transformers do. By :user:`Olivier Grisel <ogrisel>`. :pr:`33263`

- |Enhancement| :class:`kernel_approximation.Nystroem` now supports array API compatible inputs.
  By :user:`Emily Chen <EmilyXinyi>` :pr:`29661`

- |Fix| Fixed a bug that would cause Cython-based estimators to fail when fit on
  NumPy inputs when setting `sklearn.set_config(array_api_dispatch=True)`. By
  :user:`Olivier Grisel <ogrisel>`. :pr:`32846`

- |Fix| Fixes how `pos_label` is inferred when `pos_label` is set to `None`, in
  :func:`sklearn.metrics.brier_score_loss` and
  :func:`sklearn.metrics.d2_brier_score`. By :user:`Lucy Liu <lucyleeow>`. :pr:`32923`

Metadata routing
----------------

Refer to the :ref:`Metadata Routing User Guide <metadata_routing>` for
more details.

- |Enhancement| :class:`~preprocessing.TargetEncoder` now routes `groups` to the :term:`CV splitter`
  internally used for :term:`cross fitting` in its
  :meth:`~preprocessing.TargetEncoder.fit_transform`.
  By :user:`Samruddhi Baviskar <samruddhibaviskar11>` and
  :user:`Stefanie Senger <StefanieSenger>`. :pr:`33089`

:mod:`sklearn.cluster`
----------------------

- |Fix| :class:`cluster.MiniBatchKMeans` now correctly handles sample weights
  during fitting. When sample weights are not None, mini-batch
  indices are created by sub-sampling with replacement using the
  normalized sample weights as probabilities.
  By :user:`Shruti Nath <snath-xoc>`, :user:`Olivier Grisel <ogrisel>`,
  and :user:`Jeremie du Boisberranger <jeremiedbb>`. :pr:`30751`

:mod:`sklearn.compose`
----------------------

- |Fix| The dotted line for :class:`compose.ColumnTransformer` in its HTML display
  now includes only its elements. The behaviour when a remainder is used,
  has also been corrected.
  By :user:`Dea María Léon <deamarialeon>` :pr:`32713`

:mod:`sklearn.datasets`
-----------------------

- |Efficiency| Re-enabled compressed caching for :func:`datasets.fetch_kddcup99`, reducing
  on-disk cache size without changing the public API.
  By :user:`Unique Shrestha <un1u3>`. :pr:`33118`

:mod:`sklearn.ensemble`
-----------------------

- |Fix| :class:`ensemble.RandomForestClassifier`, :class:`ensemble.RandomForestRegressor`,
  :class:`ensemble.ExtraTreesClassifier` and :class:`ensemble.ExtraTreesRegressor`
  now use `sample_weight` to draw the samples instead of forwarding them
  multiplied by a uniformly sampled mask to the underlying estimators.
  Furthermore, when `max_samples` is a float, it is now interpreted as a
  fraction of `sample_weight.sum()` instead of `X.shape[0]`. As sampling is done
  with replacement, a float `max_samples` greater than `1.0` is now allowed, as
  well as an integer `max_samples` greater then `X.shape[0]`. The default
  `max_samples=None` draws `X.shape[0]` samples, irrespective of `sample_weight`.
  By :user:`Antoine Baker <antoinebaker>`. :pr:`31529`

- |Fix| Both :class:`ensemble.GradientBoostingRegressor` and
  :class:`ensemble.GradientBoostingClassifier` with the default
  `"friedman_mse"` criterion were computing impurity values with an incorrect scaling,
  leading to unexpected trees in some cases. The implementation now uses
  `"squared_error"`, which is exactly equivalent to `"friedman_mse"` up to
  floating-point error discrepancies but computes correct impurity values.
  By :user:`Arthur Lacote <cakedev0>`. :pr:`32708`

- |API| The `criterion` parameter is now deprecated for classes
  :class:`ensemble.GradientBoostingRegressor`
  and :class:`ensemble.GradientBoostingClassifier`, as both options
  (`"friedman_mse"` and `"squared_error"`) were producing the same results,
  up to floating-point rounding discrepancies and a bug in `"friedman_mse"`.
  By :user:`Arthur Lacote <cakedev0>` :pr:`32708`

:mod:`sklearn.inspection`
-------------------------

- |Fix| In :class:`inspection.DecisionBoundaryDisplay`, `multiclass_colors` is now also used
  for multiclass plotting when `response_method="predict"`.
  By :user:`Anne Beyer <AnneBeyer>`. :pr:`33015`

- |Fix| In :class:`inspection.DecisionBoundaryDisplay`, `n_classes` is now inferred more
  robustly from the estimator. If it fails for custom estimators, a comprehensive error
  message is shown.
  By :user:`Anne Beyer <AnneBeyer>`. :pr:`33202`

:mod:`sklearn.linear_model`
---------------------------

- |Efficiency| :class:`linear_model.LogisticRegression` with `solver="lbfgs"` now estimates
  the gradient of the loss at `float32` precision when fitted with `float32`
  data (`X`) to improve training speed and memory efficiency. Previously, the input
  data would be implicitly cast to `float64`. If you relied on the previous
  behavior for numerical reasons, you can explicitly cast your data to
  `float64` before fitting to reproduce it.
  By :user:`Omar Salman <OmarManzoor>` and :user:`Olivier Grisel <ogrisel>`. :pr:`32644`

- |Efficiency| The :class:`linear_model.LinearRegression`, :class:`linear_model.Ridge`,
  :class:`linear_model.Lasso`, :class:`linear_model.LassoCV`,
  :class:`linear_model.ElasticNet`, :class:`linear_model.ElasticNetCV` and
  :class:`linear_model.BayesianRidge` classes now no longer make an unnecessary copy of
  dense `X, y` input during preprocessing when `copy_X=False` and `sample_weight`
  is provided.
  By :user:`Junteng Li <JasonLiJT>`. :pr:`33041`

- |Enhancement| :class:`linear_model.ElasticNet`, :class:`linear_model.ElasticNetCV` and
  :func:`linear_model.enet_path`
  now are able to fit Ridge regression, i.e. setting `l1_ratio=0`.
  Before this PR, the stopping criterion was a formulation of the dual gap that breaks
  down for `l1_ratio=0`. Now, an alternative dual gap formulation is used for this
  setting. This reduces the noise of raised warnings.
  By :user:`Christian Lorentzen <lorentzenchr>`. :pr:`32845`

- |Enhancement| |Efficiency| :class:`linear_model.ElasticNet`, :class:`linear_model.ElasticNetCV`,
  :class:`linear_model.Lasso`, :class:`linear_model.LassoCV`,
  :class:`linear_model.MultiTaskElasticNet`, :class:`linear_model.MultiTaskElasticNetCV`
  :class:`linear_model.MultiTaskLasso`, :class:`linear_model.MultiTaskLassoCV`
  as well as
  :func:`linear_model.lasso_path` and :func:`linear_model.enet_path` are now faster when
  fit with strong L1 penalty and many features. During gap safe screening of features,
  the update of the residual is now only performed if the coefficient is not zero.
  By :user:`Christian Lorentzen <lorentzenchr>`. :pr:`33161`

- |Fix| :class:`linear_model.LassoCV` and :class:`linear_model.ElasticNetCV` now
  take the `positive` parameter into account to compute the maximum `alpha` parameter,
  where all coefficients are zero. This impacts the search grid for the
  internally tuned `alpha` hyper-parameter stored in the attribute `alphas_`.
  By :user:`Junteng Li <JasonLiJT>` :pr:`32768`

- |Fix| Correct the formulation of `alpha` within :class:`linear_model.SGDOneClassSVM`.
  The corrected value is `alpha = nu` instead of `alpha = nu / 2`.
  Note: This might result in changed values for the fitted attributes like
  `coef_` and `offset_` as well as the predictions made using this class.
  By :user:`Omar Salman <OmarManzoor>`. :pr:`32778`

- |Fix| :func:`linear_model.enet_path` now correctly handles the ``precompute`` 
  parameter when ``check_input=False``. Previously, the value of
  ``precompute`` was not properly treated which could lead to a ValueError.
  This also affects :class:`linear_model.ElasticNetCV`, :class:`linear_model.LassoCV`,
  :class:`linear_model.MultiTaskElasticNetCV` and :class:`linear_model.MultiTaskLassoCV`.
  By :user:`Albert Dorador <adc-trust-ai>` :pr:`33014`

- |Fix| Fixed a bug in :class:`linear_model.SGDClassifier` for multiclass settings where
  large negative values of :method:`decision_function` could lead to NaN values. In
  this case, this fix assigns equal probability for each class.
  By :user:`Christian Lorentzen <lorentzenchr>`. :pr:`33168`

:mod:`sklearn.metrics`
----------------------

- |Enhancement| :func:`~metrics.cohen_kappa_score` now has a `replace_undefined_by` param, that can be
  set to define the function's return value when the metric is undefined (division by
  zero).
  By :user:`Stefanie Senger <StefanieSenger>` :pr:`31172`

- |Fix| :func:`metrics.d2_pinball_score` and :func:`metrics.d2_absolute_error_score` now
  always use the `"averaged_inverted_cdf"` quantile method, both with and
  without sample weights. Previously, the `"linear"` quantile method was used only
  for the unweighted case leading the surprising discrepancies when comparing the
  results with unit weights. Note that all quantile interpolation methods are
  asymptotically equivalent in the large sample limit, but this fix can cause score
  value changes on small evaluation sets (without weights).
  By :user:`Virgil Chan <virchan>`. :pr:`31671`

:mod:`sklearn.pipeline`
-----------------------

- |Fix| Fixed :class:`pipeline.FeatureUnion` to properly handle column renaming when using Polars output, preventing duplicate column names. By :user:`Levente Csibi <leweex95>`. :pr:`32853` :pr:`32853`

:mod:`sklearn.svm`
------------------

- |Fix| Raise more informative error when fitting :class:`NuSVR` with all zero sample weights.
  By :user:`Lucy Liu <lucyleeow>` and :user:`John Hendricks <j-hendricks>`. :pr:`32212`

:mod:`sklearn.tree`
-------------------

- |Fix| Fixed feature-wise NaN detection in trees.
  Features could be seen as NaN-free for some edge-case patterns, which led to
  not considering splits with NaNs assigned to the left node for those features.
  This affects:
  - :class:`tree.DecisionTreeRegressor`
  - :class:`tree.ExtraTreeRegressor`
  - :class:`ensemble.RandomForestRegressor`
  - :class:`ensemble.ExtraTreesRegressor`
  By :user:`Arthur Lacote <cakedev0>` :pr:`32193`

- |API| `criterion="friedman_mse"` is now deprecated. This criterion was intended for
  gradient boosting but was incorrectly implemented in scikit-learn's trees and
  was actually behaving identically to `criterion="squared_error"`. Use
  `criterion="squared_error"` instead. This affects:
  - :class:`tree.DecisionTreeRegressor`
  - :class:`tree.ExtraTreeRegressor`
  - :class:`ensemble.RandomForestRegressor`
  - :class:`ensemble.ExtraTreesRegressor`
  By :user:`Arthur Lacote <cakedev0>` :pr:`32708`

:mod:`sklearn.utils`
--------------------

- |Enhancement| ``sklearn.utils._tags.get_tags`` now provides a clearer error message when a class
  is passed instead of an estimator instance.
  By :user:`Achyuthan S <Achyuthan-S>` and :user:`Anne Beyer <AnneBeyer>`. :pr:`32565`

- |Enhancement| ``sklearn.utils._response._get_response_values`` now provides a clearer error message
  when estimator does not implement the given ``response_method``.
  By :user:`Quentin Barthélemy <qbarthelemy>`. :pr:`33126`

- |Fix| The parameter table in the HTML representation of all scikit-learn
  estimators inheritiging from :class:`base.BaseEstimator`, displays
  each parameter documentation as a tooltip. The last tooltip of a
  parameter in the last table of any HTML representation was partially hidden.
  This issue has been fixed.
  By :user:`Dea María Léon <DeaMariaLeon>` :pr:`32887`

- |Fix| Fixed ``_weighted_percentile`` with ``average=True`` so zero-weight samples
   just before the end of the array are handled correctly. This
  can change results when using ``sample_weight`` with
  :class:`preprocessing.KBinsDiscretizer` (``strategy="quantile"``,
  ``quantile_method="averaged_inverted_cdf"``) and in
  :func:`metrics.median_absolute_error`, :func:`metrics.d2_pinball_score`, and
  :func:`metrics.d2_absolute_error_score`.
  By :user:`Arthur Lacote <cakedev0>`. :pr:`33127`

.. rubric:: Code and documentation contributors

Thanks to everyone who has contributed to the maintenance and improvement of
the project since version 1.8, including:

TODO: update at the time of the release.