# Release History¶

Release notes for current and recent releases are detailed on this page, with previous releases linked below.

Tip: Subscribe to scikit-learn releases on libraries.io to be notified when new versions are released.

# Version 0.21.0¶

In development

## Changed models¶

The following estimators and functions, when fit with the same data and parameters, may produce different models from the previous version. This often occurs due to changes in the modelling logic (bug fixes or enhancements), or in random sampling procedures.

• please add class and reason here (see version 0.20 what’s new)

Details are listed in the changelog below.

(While we are trying to better inform users by providing this information, we cannot assure that this list is complete.)

## Changelog¶

Support for Python 3.4 and below has been officially dropped.

• An entry goes here
• An entry goes here

## Changes to estimator checks¶

These changes mostly affect library developers.

# Version 0.20.1¶

October XX, 2018

This is a bug-fix release with some minor documentation improvements and enhancements to features released in 0.20.0.

## Changelog¶

### sklearn.linear_model¶

• Fix linear_model.SGDClassifier and variants with early_stopping=True would not use a consistent validation split in the multiclass case and this would cause a crash when using those estimators as part of parallel parameter search or cross-validation. #12122 by Olivier Grisel.

# Version 0.20.0¶

September, 2018

This release packs in a mountain of bug fixes, features and enhancements for the Scikit-learn library, and improvements to the documentation and examples. Thanks to our contributors!

This release is dedicated to the memory of Raghav Rajagopalan.

Warning

Version 0.20 is the last version of scikit-learn to support Python 2.7 and Python 3.4. Scikit-learn 0.21 will require Python 3.5 or higher.

## Highlights¶

We have tried to improve our support for common data-science use-cases including missing values, categorical variables, heterogeneous data, and features/targets with unusual distributions. Missing values in features, represented by NaNs, are now accepted in column-wise preprocessing such as scalers. Each feature is fitted disregarding NaNs, and data containing NaNs can be transformed. The new impute module provides estimators for learning despite missing data.

ColumnTransformer handles the case where different features or columns of a pandas.DataFrame need different preprocessing. String or pandas Categorical columns can now be encoded with OneHotEncoder or OrdinalEncoder.

TransformedTargetRegressor helps when the regression target needs to be transformed to be modeled. PowerTransformer and KBinsDiscretizer join QuantileTransformer as non-linear transformations.

Beyond this, we have added sample_weight support to several estimators (including KMeans, BayesianRidge and KernelDensity) and improved stopping criteria in others (including MLPRegressor, GradientBoostingRegressor and SGDRegressor).

This release is also the first to be accompanied by a Glossary of Common Terms and API Elements developed by Joel Nothman. The glossary is a reference resource to help users and contributors become familiar with the terminology and conventions used in Scikit-learn.

Sorry if your contribution didn’t make it into the highlights. There’s a lot here…

## Changed models¶

The following estimators and functions, when fit with the same data and parameters, may produce different models from the previous version. This often occurs due to changes in the modelling logic (bug fixes or enhancements), or in random sampling procedures.

Details are listed in the changelog below.

(While we are trying to better inform users by providing this information, we cannot assure that this list is complete.)

## Known Major Bugs¶

• #11924: linear_model.LogisticRegressionCV with solver='lbfgs' and multi_class='multinomial' may be non-deterministic or otherwise broken on macOS. This appears to be the case on Travis CI servers, but has not been confirmed on personal MacBooks! This issue has been present in previous releases.
• #9354: metrics.pairwise.euclidean_distances (which is used several times throughout the library) gives results with poor precision, which particularly affects its use with 32-bit float inputs. This became more problematic in versions 0.18 and 0.19 when some algorithms were changed to avoid casting 32-bit data into 64-bit.

## Changelog¶

Support for Python 3.3 has been officially dropped.

### sklearn.discriminant_analysis¶

• Efficiency Memory usage improvement for _class_means and _class_cov in discriminant_analysis. #10898 by Nanxin Chen.

### sklearn.manifold¶

• Efficiency Speed improvements for both ‘exact’ and ‘barnes_hut’ methods in manifold.TSNE. #10593 and #10610 by Tom Dupre la Tour.
• Feature Support sparse input in manifold.Isomap.fit. #8554 by Leland McInnes.
• Feature manifold.t_sne.trustworthiness accepts metrics other than Euclidean. #9775 by William de Vazelhes.
• Fix Fixed a bug in manifold.spectral_embedding where the normalization of the spectrum was using a division instead of a multiplication. #8129 by Jan Margeta, Guillaume Lemaitre, and Devansh D..
• API Change Feature Deprecate precomputed parameter in function manifold.t_sne.trustworthiness. Instead, the new parameter metric should be used with any compatible metric including ‘precomputed’, in which case the input matrix X should be a matrix of pairwise distances or squared distances. #9775 by William de Vazelhes.
• API Change Deprecate precomputed parameter in function manifold.t_sne.trustworthiness. Instead, the new parameter metric should be used with any compatible metric including ‘precomputed’, in which case the input matrix X should be a matrix of pairwise distances or squared distances. #9775 by William de Vazelhes.

### sklearn.tree¶

• Enhancement Although private (and hence not assured API stability), tree._criterion.ClassificationCriterion and tree._criterion.RegressionCriterion may now be cimported and extended. #10325 by Camil Staps.
• Fix Fixed a bug in tree.BaseDecisionTree with splitter="best" where split threshold could become infinite when values in X were near infinite. #10536 by Jonathan Ohayon.
• Fix Fixed a bug in tree.MAE to ensure sample weights are being used during the calculation of tree MAE impurity. Previous behaviour could cause suboptimal splits to be chosen since the impurity calculation considered all samples to be of equal weight importance. #11464 by John Stott.

## Changes to estimator checks¶

These changes mostly affect library developers.

## Code and Documentation Contributors¶

Thanks to everyone who has contributed to the maintenance and improvement of the project since version 0.19, including: