# 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 0.20.0¶

**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.

`cluster.MeanShift`

(bug fix)`decomposition.IncrementalPCA`

in Python 2 (bug fix)`decomposition.SparsePCA`

(bug fix)`ensemble.GradientBoostingClassifier`

(bug fix affecting feature importances)`isotonic.IsotonicRegression`

(bug fix)`linear_model.ARDRegression`

(bug fix)`linear_model.LogisticRegressionCV`

(bug fix)`linear_model.OrthogonalMatchingPursuit`

(bug fix)`linear_model.PassiveAggressiveClassifier`

(bug fix)`linear_model.PassiveAggressiveRegressor`

(bug fix)`linear_model.Perceptron`

(bug fix)`linear_model.SGDClassifier`

(bug fix)`linear_model.SGDRegressor`

(bug fix)`metrics.roc_auc_score`

(bug fix)`metrics.roc_curve`

(bug fix)`neural_network.BaseMultilayerPerceptron`

(bug fix)`neural_network.MLPClassifier`

(bug fix)`neural_network.MLPRegressor`

(bug fix)- The v0.19.0 release notes failed to mention a backwards incompatibility with
`model_selection.StratifiedKFold`

when`shuffle=True`

due to #7823.

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.cluster`

¶

- Major Feature
`cluster.AgglomerativeClustering`

now supports Single Linkage clustering via`linkage='single'`

. #9372 by Leland McInnes and Steve Astels. - Feature
`cluster.KMeans`

and`cluster.MiniBatchKMeans`

now support sample weights via new parameter`sample_weight`

in`fit`

function. #10933 by Johannes Hansen. - Efficiency
`cluster.KMeans`

,`cluster.MiniBatchKMeans`

and`cluster.k_means`

passed with`algorithm='full'`

now enforces row-major ordering, improving runtime. #10471 by Gaurav Dhingra. - Efficiency
`cluster.DBSCAN`

now is parallelized according to`n_jobs`

regardless of`algorithm`

. #8003 by Joël Billaud. - Enhancement
`cluster.KMeans`

now gives a warning if the number of distinct clusters found is smaller than`n_clusters`

. This may occur when the number of distinct points in the data set is actually smaller than the number of cluster one is looking for. #10059 by Christian Braune. - Fix Fixed a bug where the
`fit`

method of`cluster.AffinityPropagation`

stored cluster centers as 3d array instead of 2d array in case of non-convergence. For the same class, fixed undefined and arbitrary behavior in case of training data where all samples had equal similarity. #9612. By Jonatan Samoocha. - Fix Fixed a bug in
`cluster.spectral_clustering`

where the normalization of the spectrum was using a division instead of a multiplication. #8129 by Jan Margeta, Guillaume Lemaitre, and Devansh D.. - Fix Fixed a bug in
`cluster.k_means_elkan`

where the returned`iteration`

was 1 less than the correct value. Also added the missing`n_iter_`

attribute in the docstring of`cluster.KMeans`

. #11353 by Jeremie du Boisberranger. - Fix Fixed a bug in
`cluster.mean_shift`

where the assigned labels were not deterministic if there were multiple clusters with the same intensities. #11901 by Adrin Jalali. - API Change Deprecate
`pooling_func`

unused parameter in`cluster.AgglomerativeClustering`

. #9875 by Kumar Ashutosh.

`sklearn.compose`

¶

- New module.
- Major Feature Added
`compose.ColumnTransformer`

, which allows to apply different transformers to different columns of arrays or pandas DataFrames. #9012 by Andreas Müller and Joris Van den Bossche, and #11315 by Thomas Fan. - Major Feature Added the
`compose.TransformedTargetRegressor`

which transforms the target y before fitting a regression model. The predictions are mapped back to the original space via an inverse transform. #9041 by Andreas Müller and Guillaume Lemaitre.

`sklearn.covariance`

¶

- Efficiency Runtime improvements to
`covariance.GraphicalLasso`

. #9858 by Steven Brown. - API Change The
`covariance.graph_lasso`

,`covariance.GraphLasso`

and`covariance.GraphLassoCV`

have been renamed to`covariance.graphical_lasso`

,`covariance.GraphicalLasso`

and`covariance.GraphicalLassoCV`

respectively and will be removed in version 0.22. #9993 by Artiem Krinitsyn

`sklearn.datasets`

¶

- Major Feature Added
`datasets.fetch_openml`

to fetch datasets from OpenML. OpenML is a free, open data sharing platform and will be used instead of mldata as it provides better service availability. #9908 by Andreas Müller and Jan N. van Rijn. - Feature In
`datasets.make_blobs`

, one can now pass a list to the`n_samples`

parameter to indicate the number of samples to generate per cluster. #8617 by Maskani Filali Mohamed and Konstantinos Katrioplas. - Feature Add
`filename`

attribute to`datasets`

that have a CSV file. #9101 by alex-33 and Maskani Filali Mohamed. - Feature
`return_X_y`

parameter has been added to several dataset loaders. #10774 by Chris Catalfo. - Fix Fixed a bug in
`datasets.load_boston`

which had a wrong data point. #10795 by Takeshi Yoshizawa. - Fix Fixed a bug in
`datasets.load_iris`

which had two wrong data points. #11082 by Sadhana Srinivasan and Hanmin Qin. - Fix Fixed a bug in
`datasets.fetch_kddcup99`

, where data were not properly shuffled. #9731 by Nicolas Goix. - Fix Fixed a bug in
`datasets.make_circles`

, where no odd number of data points could be generated. #10045 by Christian Braune. - API Change Deprecated
`sklearn.datasets.fetch_mldata`

to be removed in version 0.22. mldata.org is no longer operational. Until removal it will remain possible to load cached datasets. #11466 by Joel Nothman.

`sklearn.decomposition`

¶

- Feature
`decomposition.dict_learning`

functions and models now support positivity constraints. This applies to the dictionary and sparse code. #6374 by John Kirkham. - Feature Fix
`decomposition.SparsePCA`

now exposes`normalize_components`

. When set to True, the train and test data are centered with the train mean repsectively during the fit phase and the transform phase. This fixes the behavior of SparsePCA. When set to False, which is the default, the previous abnormal behaviour still holds. The False value is for backward compatibility and should not be used. #11585 by Ivan Panico. - Efficiency Efficiency improvements in
`decomposition.dict_learning`

. #11420 and others by John Kirkham. - Fix Fix for uninformative error in
`decomposition.IncrementalPCA`

: now an error is raised if the number of components is larger than the chosen batch size. The`n_components=None`

case was adapted accordingly. #6452. By Wally Gauze. - Fix Fixed a bug where the
`partial_fit`

method of`decomposition.IncrementalPCA`

used integer division instead of float division on Python 2. #9492 by James Bourbeau. - Fix In
`decomposition.PCA`

selecting a n_components parameter greater than the number of samples now raises an error. Similarly, the`n_components=None`

case now selects the minimum of`n_samples`

and`n_features`

. #8484 by Wally Gauze. - Fix Fixed a bug in
`decomposition.PCA`

where users will get unexpected error with large datasets when`n_components='mle'`

on Python 3 versions. #9886 by Hanmin Qin. - Fix Fixed an underflow in calculating KL-divergence for
`decomposition.NMF`

#10142 by Tom Dupre la Tour. - Fix Fixed a bug in
`decomposition.SparseCoder`

when running OMP sparse coding in parallel using read-only memory mapped datastructures. #5956 by Vighnesh Birodkar and Olivier Grisel.

`sklearn.discriminant_analysis`

¶

- Efficiency Memory usage improvement for
`_class_means`

and`_class_cov`

in`discriminant_analysis`

. #10898 by Nanxin Chen.

`sklearn.dummy`

¶

- Feature
`dummy.DummyRegressor`

now has a`return_std`

option in its`predict`

method. The returned standard deviations will be zeros. - Feature
`dummy.DummyClassifier`

and`dummy.DummyRegressor`

now only require X to be an object with finite length or shape. #9832 by Vrishank Bhardwaj. - Feature
`dummy.DummyClassifier`

and`dummy.DummyRegressor`

can now be scored without supplying test samples. #11951 by Rüdiger Busche.

`sklearn.ensemble`

¶

- Feature
`ensemble.BaggingRegressor`

and`ensemble.BaggingClassifier`

can now be fit with missing/non-finite values in X and/or multi-output Y to support wrapping pipelines that perform their own imputation. #9707 by Jimmy Wan. - Feature
`ensemble.GradientBoostingClassifier`

and`ensemble.GradientBoostingRegressor`

now support early stopping via`n_iter_no_change`

,`validation_fraction`

and`tol`

. #7071 by Raghav RV - Feature Added
`named_estimators_`

parameter in`ensemble.VotingClassifier`

to access fitted estimators. #9157 by Herilalaina Rakotoarison. - Fix Fixed a bug when fitting
`ensemble.GradientBoostingClassifier`

or`ensemble.GradientBoostingRegressor`

with`warm_start=True`

which previously raised a segmentation fault due to a non-conversion of CSC matrix into CSR format expected by`decision_function`

. Similarly, Fortran-ordered arrays are converted to C-ordered arrays in the dense case. #9991 by Guillaume Lemaitre. - Fix Fixed a bug in
`ensemble.GradientBoostingRegressor`

and`ensemble.GradientBoostingClassifier`

to have feature importances summed and then normalized, rather than normalizing on a per-tree basis. The previous behavior over-weighted the Gini importance of features that appear in later stages. This issue only affected feature importances. #11176 by Gil Forsyth. - API Change The default value of the
`n_estimators`

parameter of`ensemble.RandomForestClassifier`

,`ensemble.RandomForestRegressor`

,`ensemble.ExtraTreesClassifier`

,`ensemble.ExtraTreesRegressor`

, and`ensemble.RandomTreesEmbedding`

will change from 10 in version 0.20 to 100 in 0.22. A FutureWarning is raised when the default value is used. #11542 by Anna Ayzenshtat. - API Change Classes derived from
`ensemble.BaseBagging`

. The attribute`estimators_samples_`

will return a list of arrays containing the indices selected for each bootstrap instead of a list of arrays containing the mask of the samples selected for each bootstrap. Indices allows to repeat samples while mask does not allow this functionality. #9524 by Guillaume Lemaitre. - Fix
`ensemble.BaseBagging`

where one could not deterministically reproduce`fit`

result using the object attributes when`random_state`

is set. #9723 by Guillaume Lemaitre.

`sklearn.feature_extraction`

¶

- Feature Enable the call to get_feature_names in unfitted
`feature_extraction.text.CountVectorizer`

initialized with a vocabulary. #10908 by Mohamed Maskani. - Enhancement
`idf_`

can now be set on a`feature_extraction.text.TfidfTransformer`

. #10899 by Sergey Melderis. - Fix Fixed a bug in
`feature_extraction.image.extract_patches_2d`

which would throw an exception if`max_patches`

was greater than or equal to the number of all possible patches rather than simply returning the number of possible patches. #10101 by Varun Agrawal - Fix Fixed a bug in
`feature_extraction.text.CountVectorizer`

,`feature_extraction.text.TfidfVectorizer`

,`feature_extraction.text.HashingVectorizer`

to support 64 bit sparse array indexing necessary to process large datasets with more than 2·10⁹ tokens (words or n-grams). #9147 by Claes-Fredrik Mannby and Roman Yurchak. - Fix Fixed bug in
`feature_extraction.text.TfidfVectorizer`

which was ignoring the parameter`dtype`

. In addition,`feature_extraction.text.TfidfTransformer`

will preserve`dtype`

for floating and raise a warning if`dtype`

requested is integer. #10441 by Mayur Kulkarni and Guillaume Lemaitre.

`sklearn.feature_selection`

¶

- Feature Added select K best features functionality to
`feature_selection.SelectFromModel`

. #6689 by Nihar Sheth and Quazi Rahman. - Feature Added
`min_features_to_select`

parameter to`feature_selection.RFECV`

to bound evaluated features counts. #11293 by Brent Yi. - Feature
`feature_selection.RFECV`

’s fit method now supports groups. #9656 by Adam Greenhall. - Fix Fixed computation of
`n_features_to_compute`

for edge case with tied CV scores in`feature_selection.RFECV`

. #9222 by Nick Hoh.

`sklearn.gaussian_process`

¶

- Efficiency In
`gaussian_process.GaussianProcessRegressor`

, method`predict`

is faster when using`return_std=True`

in particular more when called several times in a row. #9234 by andrewww and Minghui Liu.

`sklearn.impute`

¶

- New module, adopting
`preprocessing.Imputer`

as`impute.SimpleImputer`

with minor changes (see under preprocessing below). - Major Feature Added
`impute.MissingIndicator`

which generates a binary indicator for missing values. #8075 by Maniteja Nandana and Guillaume Lemaitre. - Feature The
`impute.SimpleImputer`

has a new strategy,`'constant'`

, to complete missing values with a fixed one, given by the`fill_value`

parameter. This strategy supports numeric and non-numeric data, and so does the`'most_frequent'`

strategy now. #11211 by Jeremie du Boisberranger.

`sklearn.isotonic`

¶

- Fix Fixed a bug in
`isotonic.IsotonicRegression`

which incorrectly combined weights when fitting a model to data involving points with identical X values. #9484 by Dallas Card

`sklearn.linear_model`

¶

- Feature
`linear_model.SGDClassifier`

,`linear_model.SGDRegressor`

,`linear_model.PassiveAggressiveClassifier`

,`linear_model.PassiveAggressiveRegressor`

and`linear_model.Perceptron`

now expose`early_stopping`

,`validation_fraction`

and`n_iter_no_change`

parameters, to stop optimization monitoring the score on a validation set. A new learning rate`"adaptive"`

strategy divides the learning rate by 5 each time`n_iter_no_change`

consecutive epochs fail to improve the model. #9043 by Tom Dupre la Tour. - Feature Add sample_weight parameter to the fit method of
`linear_model.BayesianRidge`

for weighted linear regression. #10112 by Peter St. John. - Fix Fixed a bug in
`logistic.logistic_regression_path`

to ensure that the returned coefficients are correct when`multiclass='multinomial'`

. Previously, some of the coefficients would override each other, leading to incorrect results in`linear_model.LogisticRegressionCV`

. #11724 by Nicolas Hug. - Fix Fixed a bug in
`linear_model.LogisticRegression`

where when using the parameter`multi_class='multinomial'`

, the`predict_proba`

method was returning incorrect probabilities in the case of binary outcomes. #9939 by Roger Westover. - Fix Fixed a bug in
`linear_model.LogisticRegressionCV`

where the`score`

method always computes accuracy, not the metric given by the`scoring`

parameter. #10998 by Thomas Fan. - Fix Fixed a bug in
`linear_model.LogisticRegressionCV`

where the ‘ovr’ strategy was always used to compute cross-validation scores in the multiclass setting, even if`'multinomial'`

was set. #8720 by William de Vazelhes. - Fix Fixed a bug in
`linear_model.OrthogonalMatchingPursuit`

that was broken when setting`normalize=False`

. #10071 by Alexandre Gramfort. - Fix Fixed a bug in
`linear_model.ARDRegression`

which caused incorrectly updated estimates for the standard deviation and the coefficients. #10153 by Jörg Döpfert. - Fix Fixed a bug in
`linear_model.ARDRegression`

and`linear_model.BayesianRidge`

which caused NaN predictions when fitted with a constant target. #10095 by Jörg Döpfert. - Fix Fixed a bug in
`linear_model.RidgeClassifierCV`

where the parameter`store_cv_values`

was not implemented though it was documented in`cv_values`

as a way to set up the storage of cross-validation values for different alphas. #10297 by Mabel Villalba-Jiménez. - Fix Fixed a bug in
`linear_model.ElasticNet`

which caused the input to be overridden when using parameter`copy_X=True`

and`check_input=False`

. #10581 by Yacine Mazari. - Fix Fixed a bug in
`sklearn.linear_model.Lasso`

where the coefficient had wrong shape when`fit_intercept=False`

. #10687 by Martin Hahn. - Fix Fixed a bug in
`sklearn.linear_model.LogisticRegression`

where the`multi_class='multinomial'`

with binary output`with warm_start=True`

#10836 by Aishwarya Srinivasan. - Fix Fixed a bug in
`linear_model.RidgeCV`

where using integer`alphas`

raised an error. #10397 by Mabel Villalba-Jiménez. - Fix Fixed condition triggering gap computation in
`linear_model.Lasso`

and`linear_model.ElasticNet`

when working with sparse matrices. #10992 by Alexandre Gramfort. - Fix Fixed a bug in
`linear_model.SGDClassifier`

,`linear_model.SGDRegressor`

,`linear_model.PassiveAggressiveClassifier`

,`linear_model.PassiveAggressiveRegressor`

and`linear_model.Perceptron`

, where the stopping criterion was stopping the algorithm before convergence. A parameter`n_iter_no_change`

was added and set by default to 5. Previous behavior is equivalent to setting the parameter to 1. #9043 by Tom Dupre la Tour. - Fix Fixed a bug where liblinear and libsvm-based estimators would segfault if passed a scipy.sparse matrix with 64-bit indices. They now raise a ValueError. #11327 by Karan Dhingra and Joel Nothman.
- API Change The default values of the
`solver`

and`multi_class`

parameters of`linear_model.LogisticRegression`

will change respectively from`'liblinear'`

and`'ovr'`

in version 0.20 to`'lbfgs'`

and`'auto'`

in version 0.22. A FutureWarning is raised when the default values are used. #11905 by Tom Dupre la Tour and Joel Nothman. - API Change Deprecate
`positive=True`

option in`linear_model.Lars`

as the underlying implementation is broken. Use`linear_model.Lasso`

instead. #9837 by Alexandre Gramfort. - API Change
`n_iter_`

may vary from previous releases in`linear_model.LogisticRegression`

with`solver='lbfgs'`

and`linear_model.HuberRegressor`

. For Scipy <= 1.0.0, the optimizer could perform more than the requested maximum number of iterations. Now both estimators will report at most`max_iter`

iterations even if more were performed. #10723 by Joel Nothman.

`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.metrics`

¶

- Major Feature Added the
`metrics.davies_bouldin_score`

metric for evaluation of clustering models without a ground truth. #10827 by Luis Osa. - Major Feature Added the
`metrics.balanced_accuracy_score`

metric and a corresponding`'balanced_accuracy'`

scorer for binary and multiclass classification. #8066 by @xyguo and Aman Dalmia, and #10587 by Joel Nothman. - Feature Partial AUC is available via
`max_fpr`

parameter in`metrics.roc_auc_score`

. #3840 by Alexander Niederbühl. - Feature A scorer based on
`metrics.brier_score_loss`

is also available. #9521 by Hanmin Qin. - Feature Added control over the normalization in
`metrics.normalized_mutual_info_score`

and`metrics.adjusted_mutual_info_score`

via the`average_method`

parameter. In version 0.22, the default normalizer for each will become the*arithmetic*mean of the entropies of each clustering. #11124 by Arya McCarthy. - Feature Added
`output_dict`

parameter in`metrics.classification_report`

to return classification statistics as dictionary. #11160 by Dan Barkhorn. - Feature
`metrics.classification_report`

now reports all applicable averages on the given data, including micro, macro and weighted average as well as samples average for multilabel data. #11679 by Alexander Pacha. - Feature
`metrics.average_precision_score`

now supports binary`y_true`

other than`{0, 1}`

or`{-1, 1}`

through`pos_label`

parameter. #9980 by Hanmin Qin. - Feature
`metrics.label_ranking_average_precision_score`

now supports`sample_weight`

. #10845 by Jose Perez-Parras Toledano. - Feature Add
`dense_output`

parameter to`metrics.pairwise.linear_kernel`

. When False and both inputs are sparse, will return a sparse matrix. #10999 by Taylor G Smith. - Efficiency
`metrics.silhouette_score`

and`metrics.silhouette_samples`

are more memory efficient and run faster. This avoids some reported freezes and MemoryErrors. #11135 by Joel Nothman. - Fix Fixed a bug in
`metrics.precision_recall_fscore_support`

when truncated range(n_labels) is passed as value for labels. #10377 by Gaurav Dhingra. - Fix Fixed a bug due to floating point error in
`metrics.roc_auc_score`

with non-integer sample weights. #9786 by Hanmin Qin. - Fix Fixed a bug where
`metrics.roc_curve`

sometimes starts on y-axis instead of (0, 0), which is inconsistent with the document and other implementations. Note that this will not influence the result from`metrics.roc_auc_score`

#10093 by alexryndin and Hanmin Qin. - Fix Fixed a bug to avoid integer overflow. Casted product to 64 bits integer in
`metrics.mutual_info_score`

. #9772 by Kumar Ashutosh. - Fix Fixed a bug where
`metrics.average_precision_score`

will sometimes return`nan`

when`sample_weight`

contains 0. #9980 by Hanmin Qin. - Fix Fixed a bug in
`metrics.fowlkes_mallows_score`

to avoid integer overflow. Casted return value of contingency_matrix to int64 and computed product of square roots rather than square root of product. #9515 by Alan Liddell and Manh Dao. - API Change Deprecate
`reorder`

parameter in`metrics.auc`

as it’s no longer required for`metrics.roc_auc_score`

. Moreover using`reorder=True`

can hide bugs due to floating point error in the input. #9851 by Hanmin Qin. - API Change In
`metrics.normalized_mutual_info_score`

and`metrics.adjusted_mutual_info_score`

, warn that`average_method`

will have a new default value. In version 0.22, the default normalizer for each will become the*arithmetic*mean of the entropies of each clustering. Currently,`metrics.normalized_mutual_info_score`

uses the default of`average_method='geometric'`

, and`metrics.adjusted_mutual_info_score`

uses the default of`average_method='max'`

to match their behaviors in version 0.19. #11124 by Arya McCarthy. - API Change The
`batch_size`

parameter to`metrics.pairwise_distances_argmin_min`

and`metrics.pairwise_distances_argmin`

is deprecated to be removed in v0.22. It no longer has any effect, as batch size is determined by global`working_memory`

config. See Limiting Working Memory. #10280 by Joel Nothman and Aman Dalmia.

`sklearn.mixture`

¶

- Feature Added function fit_predict to
`mixture.GaussianMixture`

and`mixture.GaussianMixture`

, which is essentially equivalent to calling fit and predict. #10336 by Shu Haoran and Andrew Peng. - Fix Fixed a bug in
`mixture.BaseMixture`

where the reported n_iter_ was missing an iteration. It affected`mixture.GaussianMixture`

and`mixture.BayesianGaussianMixture`

. #10740 by Erich Schubert and Guillaume Lemaitre. - Fix Fixed a bug in
`mixture.BaseMixture`

and its subclasses`mixture.GaussianMixture`

and`mixture.BayesianGaussianMixture`

where the`lower_bound_`

was not the max lower bound across all initializations (when`n_init > 1`

), but just the lower bound of the last initialization. #10869 by Aurélien Géron.

`sklearn.model_selection`

¶

- Feature Add return_estimator parameter in
`model_selection.cross_validate`

to return estimators fitted on each split. #9686 by Aurélien Bellet. - Feature New
`refit_time_`

attribute will be stored in`model_selection.GridSearchCV`

and`model_selection.RandomizedSearchCV`

if`refit`

is set to`True`

. This will allow measuring the complete time it takes to perform hyperparameter optimization and refitting the best model on the whole dataset. #11310 by Matthias Feurer. - Feature Expose error_score parameter in
`model_selection.cross_validate`

,`model_selection.cross_val_score`

,`model_selection.learning_curve`

and`model_selection.validation_curve`

to control the behavior triggered when an error occurs in`model_selection._fit_and_score`

. #11576 by Samuel O. Ronsin. - Feature BaseSearchCV now has an experimental, private interface to
support customized parameter search strategies, through its
`_run_search`

method. See the implementations in`model_selection.GridSearchCV`

and`model_selection.RandomizedSearchCV`

and please provide feedback if you use this. Note that we do not assure the stability of this API beyond version 0.20. #9599 by Joel Nothman - Enhancement Add improved error message in
`model_selection.cross_val_score`

when multiple metrics are passed in`scoring`

keyword. #11006 by Ming Li. - API Change The default number of cross-validation folds
`cv`

and the default number of splits`n_splits`

in the`model_selection.KFold`

-like splitters will change from 3 to 5 in 0.22 as 3-fold has a lot of variance. #11557 by Alexandre Boucaud. - API Change The default of
`iid`

parameter of`model_selection.GridSearchCV`

and`model_selection.RandomizedSearchCV`

will change from`True`

to`False`

in version 0.22 to correspond to the standard definition of cross-validation, and the parameter will be removed in version 0.24 altogether. This parameter is of greatest practical significance where the sizes of different test sets in cross-validation were very unequal, i.e. in group-based CV strategies. #9085 by Laurent Direr and Andreas Müller. - API Change The default value of the
`error_score`

parameter in`model_selection.GridSearchCV`

and`model_selection.RandomizedSearchCV`

will change to`np.NaN`

in version 0.22. #10677 by Kirill Zhdanovich. - API Change Changed ValueError exception raised in
`model_selection.ParameterSampler`

to a UserWarning for case where the class is instantiated with a greater value of`n_iter`

than the total space of parameters in the parameter grid.`n_iter`

now acts as an upper bound on iterations. #10982 by Juliet Lawton - API Change Invalid input for
`model_selection.ParameterGrid`

now raises TypeError. #10928 by Solutus Immensus

`sklearn.multioutput`

¶

- Major Feature Added
`multioutput.RegressorChain`

for multi-target regression. #9257 by Kumar Ashutosh.

`sklearn.naive_bayes`

¶

- Major Feature Added
`naive_bayes.ComplementNB`

, which implements the Complement Naive Bayes classifier described in Rennie et al. (2003). #8190 by Michael A. Alcorn. - Feature Add var_smoothing parameter in
`naive_bayes.GaussianNB`

to give a precise control over variances calculation. #9681 by Dmitry Mottl. - Fix Fixed a bug in
`naive_bayes.GaussianNB`

which incorrectly raised error for prior list which summed to 1. #10005 by Gaurav Dhingra. - Fix Fixed a bug in
`naive_bayes.MultinomialNB`

which did not accept vector valued pseudocounts (alpha). #10346 by Tobias Madsen

`sklearn.neighbors`

¶

- Efficiency
`neighbors.RadiusNeighborsRegressor`

and`neighbors.RadiusNeighborsClassifier`

are now parallelized according to`n_jobs`

regardless of`algorithm`

. #10887 by Joël Billaud. - Efficiency
`Nearest neighbors`

query methods are now more memory efficient when`algorithm='brute'`

. #11136 by Joel Nothman and Aman Dalmia. - Feature Add
`sample_weight`

parameter to the fit method of`neighbors.KernelDensity`

to enable weighting in kernel density estimation. #4394 by Samuel O. Ronsin. - Feature Novelty detection with
`neighbors.LocalOutlierFactor`

: Add a`novelty`

parameter to`neighbors.LocalOutlierFactor`

. When`novelty`

is set to True,`neighbors.LocalOutlierFactor`

can then be used for novelty detection, i.e. predict on new unseen data. Available prediction methods are`predict`

,`decision_function`

and`score_samples`

. By default,`novelty`

is set to`False`

, and only the`fit_predict`

method is avaiable. By Albert Thomas. - Fix Fixed a bug in
`neighbors.NearestNeighbors`

where fitting a NearestNeighbors model fails when a) the distance metric used is a callable and b) the input to the NearestNeighbors model is sparse. #9579 by Thomas Kober. - Fix Fixed a bug so
`predict`

in`neighbors.RadiusNeighborsRegressor`

can handle empty neighbor set when using non uniform weights. Also raises a new warning when no neighbors are found for samples. #9655 by Andreas Bjerre-Nielsen. - Fix Efficiency Fixed a bug in
`KDTree`

construction that results in faster construction and querying times. #11556 by Jake VanderPlas - Fix Fixed a bug in
`neighbors.KDTree`

and`neighbors.BallTree`

where pickled tree objects would change their type to the super class`BinaryTree`

. #11774 by Nicolas Hug.

`sklearn.neural_network`

¶

- Feature Add n_iter_no_change parameter in
`neural_network.BaseMultilayerPerceptron`

,`neural_network.MLPRegressor`

, and`neural_network.MLPClassifier`

to give control over maximum number of epochs to not meet`tol`

improvement. #9456 by Nicholas Nadeau. - Fix Fixed a bug in
`neural_network.BaseMultilayerPerceptron`

,`neural_network.MLPRegressor`

, and`neural_network.MLPClassifier`

with new`n_iter_no_change`

parameter now at 10 from previously hardcoded 2. #9456 by Nicholas Nadeau. - Fix Fixed a bug in
`neural_network.MLPRegressor`

where fitting quit unexpectedly early due to local minima or fluctuations. #9456 by Nicholas Nadeau

`sklearn.pipeline`

¶

- Feature The
`predict`

method of`pipeline.Pipeline`

now passes keyword arguments on to the pipeline’s last estimator, enabling the use of parameters such as`return_std`

in a pipeline with caution. #9304 by Breno Freitas. - API Change
`pipeline.FeatureUnion`

now supports`'drop'`

as a transformer to drop features. #11144 by @thomasjpfan.

`sklearn.preprocessing`

¶

- Major Feature Expanded
`preprocessing.OneHotEncoder`

to allow to encode categorical string features as a numeric array using a one-hot (or dummy) encoding scheme, and added`preprocessing.OrdinalEncoder`

to convert to ordinal integers. Those two classes now handle encoding of all feature types (also handles string-valued features) and derives the categories based on the unique values in the features instead of the maximum value in the features. #9151 and #10521 by Vighnesh Birodkar and Joris Van den Bossche. - Major Feature Added
`preprocessing.KBinsDiscretizer`

for turning continuous features into categorical or one-hot encoded features. #7668, #9647, #10195, #10192, #11272, #11467 and #11505. by Henry Lin, Hanmin Qin, Tom Dupre la Tour and Giovanni Giuseppe Costa. - Major Feature Added
`preprocessing.PowerTransformer`

, which implements the Yeo-Johnson and Box-Cox power transformations. Power transformations try to find a set of feature-wise parametric transformations to approximately map data to a Gaussian distribution centered at zero and with unit variance. This is useful as a variance-stabilizing transformation in situations where normality and homoscedasticity are desirable. #10210 by Eric Chang and Maniteja Nandana, and #11520 by Nicolas Hug. - Major Feature NaN values are ignored and handled in the following
preprocessing methods:
`preprocessing.MaxAbsScaler`

,`preprocessing.MinMaxScaler`

,`preprocessing.RobustScaler`

,`preprocessing.StandardScaler`

,`preprocessing.PowerTransformer`

,`preprocessing.QuantileTransformer`

classes and`preprocessing.maxabs_scale`

,`preprocessing.minmax_scale`

,`preprocessing.robust_scale`

,`preprocessing.scale`

,`preprocessing.power_transform`

,`preprocessing.quantile_transform`

functions respectively addressed in issues #11011, #11005, #11308, #11206, #11306, and #10437. By Lucija Gregov and Guillaume Lemaitre. - Feature
`preprocessing.PolynomialFeatures`

now supports sparse input. #10452 by Aman Dalmia and Joel Nothman. - Feature
`preprocessing.RobustScaler`

and`preprocessing.robust_scale`

can be fitted using sparse matrices. #11308 by Guillaume Lemaitre. - Feature
`preprocessing.OneHotEncoder`

now supports the get_feature_names method to obtain the transformed feature names. #10181 by Nirvan Anjirbag and Joris Van den Bossche. - Feature A parameter
`check_inverse`

was added to`preprocessing.FunctionTransformer`

to ensure that`func`

and`inverse_func`

are the inverse of each other. #9399 by Guillaume Lemaitre. - Feature The
`transform`

method of`sklearn.preprocessing.MultiLabelBinarizer`

now ignores any unknown classes. A warning is raised stating the unknown classes classes found which are ignored. #10913 by Rodrigo Agundez. - Fix Fixed bugs in
`preprocessing.LabelEncoder`

which would sometimes throw errors when`transform`

or`inverse_transform`

was called with empty arrays. #10458 by Mayur Kulkarni. - Fix Fix ValueError in
`preprocessing.LabelEncoder`

when using`inverse_transform`

on unseen labels. #9816 by Charlie Newey. - Fix Fix bug in
`preprocessing.OneHotEncoder`

which discarded the`dtype`

when returning a sparse matrix output. #11042 by Daniel Morales. - Fix Fix
`fit`

and`partial_fit`

in`preprocessing.StandardScaler`

in the rare case when`with_mean=False`

and with_std=False which was crashing by calling`fit`

more than once and giving inconsistent results for`mean_`

whether the input was a sparse or a dense matrix.`mean_`

will be set to`None`

with both sparse and dense inputs.`n_samples_seen_`

will be also reported for both input types. #11235 by Guillaume Lemaitre. - API Change Deprecate
`n_values`

and`categorical_features`

parameters and`active_features_`

,`feature_indices_`

and`n_values_`

attributes of`preprocessing.OneHotEncoder`

. The`n_values`

parameter can be replaced with the new`categories`

parameter, and the attributes with the new`categories_`

attribute. Selecting the categorical features with the`categorical_features`

parameter is now better supported using the`compose.ColumnTransformer`

. #10521 by Joris Van den Bossche. - API Change Deprecate
`preprocessing.Imputer`

and move the corresponding module to`impute.SimpleImputer`

. #9726 by Kumar Ashutosh. - API Change The
`axis`

parameter that was in`preprocessing.Imputer`

is no longer present in`impute.SimpleImputer`

. The behavior is equivalent to`axis=0`

(impute along columns). Row-wise imputation can be performed with FunctionTransformer (e.g.,`FunctionTransformer(lambda X: SimpleImputer().fit_transform(X.T).T)`

). #10829 by Guillaume Lemaitre and Gilberto Olimpio. - API Change The NaN marker for the missing values has been changed
between the
`preprocessing.Imputer`

and the`impute.SimpleImputer`

.`missing_values='NaN'`

should now be`missing_values=np.nan`

. #11211 by Jeremie du Boisberranger. - API Change In
`preprocessing.FunctionTransformer`

, the default of`validate`

will be from`True`

to`False`

in 0.22. #10655 by Guillaume Lemaitre.

`sklearn.svm`

¶

- Fix Fixed a bug in
`svm.SVC`

where when the argument`kernel`

is unicode in Python2, the`predict_proba`

method was raising an unexpected TypeError given dense inputs. #10412 by Jiongyan Zhang. - API Change Deprecate
`random_state`

parameter in`svm.OneClassSVM`

as the underlying implementation is not random. #9497 by Albert Thomas. - API Change The default value of
`gamma`

parameter of`svm.SVC`

,`NuSVC`

,`SVR`

,`NuSVR`

,`OneClassSVM`

will change from`'auto'`

to`'scale'`

in version 0.22 to account better for unscaled features. #8361 by Gaurav Dhingra and Ting Neo.

`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.

`sklearn.utils`

¶

- Feature
`utils.check_array`

and`utils.check_X_y`

now have`accept_large_sparse`

to control whether scipy.sparse matrices with 64-bit indices should be rejected. #11327 by Karan Dhingra and Joel Nothman. - Efficiency Fix Avoid copying the data in
`utils.check_array`

when the input data is a memmap (and`copy=False`

). #10663 by Arthur Mensch and Loïc Estève. - API Change
`utils.check_array`

yield a`FutureWarning`

indicating that arrays of bytes/strings will be interpreted as decimal numbers beginning in version 0.22. #10229 by Ryan Lee

### Multiple modules¶

- Feature API Change More consistent outlier detection API:
Add a
`score_samples`

method in`svm.OneClassSVM`

,`ensemble.IsolationForest`

,`neighbors.LocalOutlierFactor`

,`covariance.EllipticEnvelope`

. It allows to access raw score functions from original papers. A new`offset_`

parameter allows to link`score_samples`

and`decision_function`

methods. The`contamination`

parameter of`ensemble.IsolationForest`

and`neighbors.LocalOutlierFactor`

`decision_function`

methods is used to define this`offset_`

such that outliers (resp. inliers) have negative (resp. positive)`decision_function`

values. By default,`contamination`

is kept unchanged to 0.1 for a deprecation period. In 0.22, it will be set to “auto”, thus using method-specific score offsets. In`covariance.EllipticEnvelope`

`decision_function`

method, the`raw_values`

parameter is deprecated as the shifted Mahalanobis distance will be always returned in 0.22. #9015 by Nicolas Goix. - Feature API Change A
`behaviour`

parameter has been introduced in`ensemble.IsolationForest`

to ensure backward compatibility. In the old behaviour, the`decision_function`

is independent of the`contamination`

parameter. A threshold attribute depending on the`contamination`

parameter is thus used. In the new behaviour the`decision_function`

is dependent on the`contamination`

parameter, in such a way that 0 becomes its natural threshold to detect outliers. Setting behaviour to “old” is deprecated and will not be possible in version 0.22. Beside, the behaviour parameter will be removed in 0.24. #11553 by Nicolas Goix. - API Change Added convergence warning to
`svm.LinearSVC`

and`linear_model.LogisticRegression`

when`verbose`

is set to 0. #10881 by Alexandre Sevin. - API Change Changed warning type from
`UserWarning`

to`exceptions.ConvergenceWarning`

for failing convergence in`linear_model.logistic_regression_path`

,`linear_model.RANSACRegressor`

,`linear_model.ridge_regression`

,`gaussian_process.GaussianProcessRegressor`

,`gaussian_process.GaussianProcessClassifier`

,`decomposition.fastica`

,`cross_decomposition.PLSCanonical`

,`cluster.AffinityPropagation`

, and`cluster.Birch`

. #10306 by Jonathan Siebert.

### Miscellaneous¶

- Major Feature A new configuration parameter,
`working_memory`

was added to control memory consumption limits in chunked operations, such as the new`metrics.pairwise_distances_chunked`

. See Limiting Working Memory. #10280 by Joel Nothman and Aman Dalmia. - Feature The version of
`joblib`

bundled with Scikit-learn is now 0.12. This uses a new default multiprocessing implementation, named loky. While this may incur some memory and communication overhead, it should provide greater cross-platform stability than relying on Python standard library multiprocessing. #11741 by the Joblib developers, especially Thomas Moreau and Olivier Grisel. - Feature An environment variable to use the site joblib instead of the
vendored one was added (Environment variables). The main API of joblib
is now exposed in
`sklearn.utils`

. #11166 by Gael Varoquaux. - Feature Add almost complete PyPy 3 support. Known unsupported
functionalities are
`datasets.load_svmlight_file`

,`feature_extraction.FeatureHasher`

and`feature_extraction.text.HashingVectorizer`

. For running on PyPy, PyPy3-v5.10+, Numpy 1.14.0+, and scipy 1.1.0+ are required. #11010 by Ronan Lamy and Roman Yurchak. - Feature A utility method
`sklearn.show_versions`

was added to print out information relevant for debugging. It includes the user system, the Python executable, the version of the main libraries and BLAS binding information. #11596 by Alexandre Boucaud - Fix Fixed a bug when setting parameters on meta-estimator, involving both a wrapped estimator and its parameter. #9999 by Marcus Voss and Joel Nothman.
- Fix Fixed a bug where calling
`sklearn.base.clone`

was not thread safe and could result in a “pop from empty list” error. #9569 by Andreas Müller. - API Change The default value of
`n_jobs`

is changed from`1`

to`None`

in all related functions and classes.`n_jobs=None`

means`unset`

. It will generally be interpreted as`n_jobs=1`

, unless the current`joblib.Parallel`

backend context specifies otherwise (See Glossary for additional information). Note that this change happens immediately (i.e., without a deprecation cycle). #11741 by Olivier Grisel.

## Changes to estimator checks¶

These changes mostly affect library developers.

- Checks for transformers now apply if the estimator implements
transform, regardless of whether it inherits from
`sklearn.base.TransformerMixin`

. #10474 by Joel Nothman. - Classifiers are now checked for consistency between decision_function and categorical predictions. #10500 by Narine Kokhlikyan.
- Allow tests in
`utils.estimator_checks.check_estimator`

to test functions that accept pairwise data. #9701 by Kyle Johnson - Allow
`utils.estimator_checks.check_estimator`

to check that there is no private settings apart from parameters during estimator initialization. #9378 by Herilalaina Rakotoarison - The set of checks in
`utils.estimator_checks.check_estimator`

now includes a`check_set_params`

test which checks that`set_params`

is equivalent to passing parameters in`__init__`

and warns if it encounters parameter validation. #7738 by Alvin Chiang - Add invariance tests for clustering metrics. #8102 by Ankita Sinha and Guillaume Lemaitre.
- Add
`check_methods_subset_invariance`

to`check_estimator`

, which checks that estimator methods are invariant if applied to a data subset. #10428 by Jonathan Ohayon - Add tests in
`utils.estimator_checks.check_estimator`

to check that an estimator can handle read-only memmap input data. #10663 by Arthur Mensch and Loïc Estève. `check_sample_weights_pandas_series`

now uses 8 rather than 6 samples to accommodate for the default number of clusters in`cluster.KMeans`

. #10933 by Johannes Hansen.- Estimators are now checked for whether
`sample_weight=None`

equates to`sample_weight=np.ones(...)`

. #11558 by Sergul Aydore.

## Code and Documentation Contributors¶

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

211217613, Aarshay Jain, absolutelyNoWarranty, Adam Greenhall, Adam Kleczewski, Adam Richie-Halford, adelr, AdityaDaflapurkar, Adrin Jalali, Aidan Fitzgerald, aishgrt1, Akash Shivram, Alan Liddell, Alan Yee, Albert Thomas, Alexander Lenail, Alexander-N, Alexandre Boucaud, Alexandre Gramfort, Alexandre Sevin, Alex Egg, Alvaro Perez-Diaz, Amanda, Aman Dalmia, Andreas Bjerre-Nielsen, Andreas Mueller, Andrew Peng, Angus Williams, Aniruddha Dave, annaayzenshtat, Anthony Gitter, Antonio Quinonez, Anubhav Marwaha, Arik Pamnani, Arthur Ozga, Artiem K, Arunava, Arya McCarthy, Attractadore, Aurélien Bellet, Aurélien Geron, Ayush Gupta, Balakumaran Manoharan, Bangda Sun, Barry Hart, Bastian Venthur, Ben Lawson, Benn Roth, Breno Freitas, Brent Yi, brett koonce, Caio Oliveira, Camil Staps, cclauss, Chady Kamar, Charlie Brummitt, Charlie Newey, chris, Chris, Chris Catalfo, Chris Foster, Chris Holdgraf, Christian Braune, Christian Hirsch, Christian Hogan, Christopher Jenness, Clement Joudet, cnx, cwitte, Dallas Card, Dan Barkhorn, Daniel, Daniel Ferreira, Daniel Gomez, Daniel Klevebring, Danielle Shwed, Daniel Mohns, Danil Baibak, Darius Morawiec, David Beach, David Burns, David Kirkby, David Nicholson, David Pickup, Derek, Didi Bar-Zev, diegodlh, Dillon Gardner, Dillon Niederhut, dilutedsauce, dlovell, Dmitry Mottl, Dmitry Petrov, Dor Cohen, Douglas Duhaime, Ekaterina Tuzova, Eric Chang, Eric Dean Sanchez, Erich Schubert, Eunji, Fang-Chieh Chou, FarahSaeed, felix, Félix Raimundo, fenx, filipj8, FrankHui, Franz Wompner, Freija Descamps, frsi, Gabriele Calvo, Gael Varoquaux, Gaurav Dhingra, Georgi Peev, Gil Forsyth, Giovanni Giuseppe Costa, gkevinyen5418, goncalo-rodrigues, Gryllos Prokopis, Guillaume Lemaitre, Guillaume “Vermeille” Sanchez, Gustavo De Mari Pereira, hakaa1, Hanmin Qin, Henry Lin, Hong, Honghe, Hossein Pourbozorg, Hristo, Hunan Rostomyan, iampat, Ivan PANICO, Jaewon Chung, Jake VanderPlas, jakirkham, James Bourbeau, James Malcolm, Jamie Cox, Jan Koch, Jan Margeta, Jan Schlüter, janvanrijn, Jason Wolosonovich, JC Liu, Jeb Bearer, jeremiedbb, Jimmy Wan, Jinkun Wang, Jiongyan Zhang, jjabl, jkleint, Joan Massich, Joël Billaud, Joel Nothman, Johannes Hansen, JohnStott, Jonatan Samoocha, Jonathan Ohayon, Jörg Döpfert, Joris Van den Bossche, Jose Perez-Parras Toledano, josephsalmon, jotasi, jschendel, Julian Kuhlmann, Julien Chaumond, julietcl, Justin Shenk, Karl F, Kasper Primdal Lauritzen, Katrin Leinweber, Kirill, ksemb, Kuai Yu, Kumar Ashutosh, Kyeongpil Kang, Kye Taylor, kyledrogo, Leland McInnes, Léo DS, Liam Geron, Liutong Zhou, Lizao Li, lkjcalc, Loic Esteve, louib, Luciano Viola, Lucija Gregov, Luis Osa, Luis Pedro Coelho, Luke M Craig, Luke Persola, Mabel, Mabel Villalba, Maniteja Nandana, MarkIwanchyshyn, Mark Roth, Markus Müller, MarsGuy, Martin Gubri, martin-hahn, martin-kokos, mathurinm, Matthias Feurer, Max Copeland, Mayur Kulkarni, Meghann Agarwal, Melanie Goetz, Michael A. Alcorn, Minghui Liu, Ming Li, Minh Le, Mohamed Ali Jamaoui, Mohamed Maskani, Mohammad Shahebaz, Muayyad Alsadi, Nabarun Pal, Nagarjuna Kumar, Naoya Kanai, Narendran Santhanam, NarineK, Nathaniel Saul, Nathan Suh, Nicholas Nadeau, P.Eng., AVS, Nick Hoh, Nicolas Goix, Nicolas Hug, Nicolau Werneck, nielsenmarkus11, Nihar Sheth, Nikita Titov, Nilesh Kevlani, Nirvan Anjirbag, notmatthancock, nzw, Oleksandr Pavlyk, oliblum90, Oliver Rausch, Olivier Grisel, Oren Milman, Osaid Rehman Nasir, pasbi, Patrick Fernandes, Patrick Olden, Paul Paczuski, Pedro Morales, Peter, Peter St. John, pierreablin, pietruh, Pinaki Nath Chowdhury, Piotr Szymański, Pradeep Reddy Raamana, Pravar D Mahajan, pravarmahajan, QingYing Chen, Raghav RV, Rajendra arora, RAKOTOARISON Herilalaina, Rameshwar Bhaskaran, RankyLau, Rasul Kerimov, Reiichiro Nakano, Rob, Roman Kosobrodov, Roman Yurchak, Ronan Lamy, rragundez, Rüdiger Busche, Ryan, Sachin Kelkar, Sagnik Bhattacharya, Sailesh Choyal, Sam Radhakrishnan, Sam Steingold, Samuel Bell, Samuel O. Ronsin, Saqib Nizam Shamsi, SATISH J, Saurabh Gupta, Scott Gigante, Sebastian Flennerhag, Sebastian Raschka, Sebastien Dubois, Sébastien Lerique, Sebastin Santy, Sergey Feldman, Sergey Melderis, Sergul Aydore, Shahebaz, Shalil Awaley, Shangwu Yao, Sharad Vijalapuram, Sharan Yalburgi, shenhanc78, Shivam Rastogi, Shu Haoran, siftikha, Sinclert Pérez, SolutusImmensus, Somya Anand, srajan paliwal, Sriharsha Hatwar, Sri Krishna, Stefan van der Walt, Stephen McDowell, Steven Brown, syonekura, Taehoon Lee, Takanori Hayashi, tarcusx, Taylor G Smith, theriley106, Thomas, Thomas Fan, Thomas Heavey, Tobias Madsen, tobycheese, Tom Augspurger, Tom Dupré la Tour, Tommy, Trevor Stephens, Trishnendu Ghorai, Tulio Casagrande, twosigmajab, Umar Farouk Umar, Urvang Patel, Utkarsh Upadhyay, Vadim Markovtsev, Varun Agrawal, Vathsala Achar, Vilhelm von Ehrenheim, Vinayak Mehta, Vinit, Vinod Kumar L, Viraj Mavani, Viraj Navkal, Vivek Kumar, Vlad Niculae, vqean3, Vrishank Bhardwaj, vufg, wallygauze, Warut Vijitbenjaronk, wdevazelhes, Wenhao Zhang, Wes Barnett, Will, William de Vazelhes, Will Rosenfeld, Xin Xiong, Yiming (Paul) Li, ymazari, Yufeng, Zach Griffith, Zé Vinícius, Zhenqing Hu, Zhiqing Xiao, Zijie (ZJ) Poh

# Version 0.19.1¶

**October 23, 2017**

This is a bug-fix release with some minor documentation improvements and enhancements to features released in 0.19.0.

Note there may be minor differences in TSNE output in this release (due to #9623), in the case where multiple samples have equal distance to some sample.

## Changelog¶

### API changes¶

- Reverted the addition of
`metrics.ndcg_score`

and`metrics.dcg_score`

which had been merged into version 0.19.0 by error. The implementations were broken and undocumented. `return_train_score`

which was added to`model_selection.GridSearchCV`

,`model_selection.RandomizedSearchCV`

and`model_selection.cross_validate`

in version 0.19.0 will be changing its default value from True to False in version 0.21. We found that calculating training score could have a great effect on cross validation runtime in some cases. Users should explicitly set`return_train_score`

to False if prediction or scoring functions are slow, resulting in a deleterious effect on CV runtime, or to True if they wish to use the calculated scores. #9677 by Kumar Ashutosh and Joel Nothman.`correlation_models`

and`regression_models`

from the legacy gaussian processes implementation have been belatedly deprecated. #9717 by Kumar Ashutosh.

### Bug fixes¶

- Avoid integer overflows in
`metrics.matthews_corrcoef`

. #9693 by Sam Steingold. - Fixed a bug in the objective function for
`manifold.TSNE`

(both exact and with the Barnes-Hut approximation) when`n_components >= 3`

. #9711 by @goncalo-rodrigues. - Fix regression in
`model_selection.cross_val_predict`

where it raised an error with`method='predict_proba'`

for some probabilistic classifiers. #9641 by James Bourbeau. - Fixed a bug where
`datasets.make_classification`

modified its input`weights`

. #9865 by Sachin Kelkar. `model_selection.StratifiedShuffleSplit`

now works with multioutput multiclass or multilabel data with more than 1000 columns. #9922 by Charlie Brummitt.- Fixed a bug with nested and conditional parameter setting, e.g. setting a pipeline step and its parameter at the same time. #9945 by Andreas Müller and Joel Nothman.

Regressions in 0.19.0 fixed in 0.19.1:

- Fixed a bug where parallelised prediction in random forests was not thread-safe and could (rarely) result in arbitrary errors. #9830 by Joel Nothman.
- Fix regression in
`model_selection.cross_val_predict`

where it no longer accepted`X`

as a list. #9600 by Rasul Kerimov. - Fixed handling of
`cross_val_predict`

for binary classification with`method='decision_function'`

. #9593 by Reiichiro Nakano and core devs. - Fix regression in
`pipeline.Pipeline`

where it no longer accepted`steps`

as a tuple. #9604 by Joris Van den Bossche. - Fix bug where
`n_iter`

was not properly deprecated, leaving`n_iter`

unavailable for interim use in`linear_model.SGDClassifier`

,`linear_model.SGDRegressor`

,`linear_model.PassiveAggressiveClassifier`

,`linear_model.PassiveAggressiveRegressor`

and`linear_model.Perceptron`

. #9558 by Andreas Müller. - Dataset fetchers make sure temporary files are closed before removing them, which caused errors on Windows. #9847 by Joan Massich.
- Fixed a regression in
`manifold.TSNE`

where it no longer supported metrics other than ‘euclidean’ and ‘precomputed’. #9623 by Oli Blum.

### Enhancements¶

- Our test suite and
`utils.estimator_checks.check_estimators`

can now be run without Nose installed. #9697 by Joan Massich. - To improve usability of version 0.19’s
`pipeline.Pipeline`

caching,`memory`

now allows`joblib.Memory`

instances. This make use of the new`utils.validation.check_memory`

helper. issue:9584 by Kumar Ashutosh - Some fixes to examples: #9750, #9788, #9815
- Made a FutureWarning in SGD-based estimators less verbose. #9802 by Vrishank Bhardwaj.

## Code and Documentation Contributors¶

With thanks to:

Joel Nothman, Loic Esteve, Andreas Mueller, Kumar Ashutosh, Vrishank Bhardwaj, Hanmin Qin, Rasul Kerimov, James Bourbeau, Nagarjuna Kumar, Nathaniel Saul, Olivier Grisel, Roman Yurchak, Reiichiro Nakano, Sachin Kelkar, Sam Steingold, Yaroslav Halchenko, diegodlh, felix, goncalo-rodrigues, jkleint, oliblum90, pasbi, Anthony Gitter, Ben Lawson, Charlie Brummitt, Didi Bar-Zev, Gael Varoquaux, Joan Massich, Joris Van den Bossche, nielsenmarkus11

# Version 0.19¶

**August 12, 2017**

## Highlights¶

We are excited to release a number of great new features including
`neighbors.LocalOutlierFactor`

for anomaly detection,
`preprocessing.QuantileTransformer`

for robust feature transformation,
and the `multioutput.ClassifierChain`

meta-estimator to simply account
for dependencies between classes in multilabel problems. We have some new
algorithms in existing estimators, such as multiplicative update in
`decomposition.NMF`

and multinomial
`linear_model.LogisticRegression`

with L1 loss (use `solver='saga'`

).

Cross validation is now able to return the results from multiple metric
evaluations. The new `model_selection.cross_validate`

can return many
scores on the test data as well as training set performance and timings, and we
have extended the `scoring`

and `refit`

parameters for grid/randomized
search to handle multiple metrics.

You can also learn faster. For instance, the new option to cache
transformations in `pipeline.Pipeline`

makes grid
search over pipelines including slow transformations much more efficient. And
you can predict faster: if you’re sure you know what you’re doing, you can turn
off validating that the input is finite using `config_context`

.

We’ve made some important fixes too. We’ve fixed a longstanding implementation
error in `metrics.average_precision_score`

, so please be cautious with
prior results reported from that function. A number of errors in the
`manifold.TSNE`

implementation have been fixed, particularly in the
default Barnes-Hut approximation. `semi_supervised.LabelSpreading`

and
`semi_supervised.LabelPropagation`

have had substantial fixes.
LabelPropagation was previously broken. LabelSpreading should now correctly
respect its alpha parameter.

## 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.

`cluster.KMeans`

with sparse X and initial centroids given (bug fix)`cross_decomposition.PLSRegression`

with`scale=True`

(bug fix)`ensemble.GradientBoostingClassifier`

and`ensemble.GradientBoostingRegressor`

where`min_impurity_split`

is used (bug fix)- gradient boosting
`loss='quantile'`

(bug fix) `ensemble.IsolationForest`

(bug fix)`feature_selection.SelectFdr`

(bug fix)`linear_model.RANSACRegressor`

(bug fix)`linear_model.LassoLars`

(bug fix)`linear_model.LassoLarsIC`

(bug fix)`manifold.TSNE`

(bug fix)`neighbors.NearestCentroid`

(bug fix)`semi_supervised.LabelSpreading`

(bug fix)`semi_supervised.LabelPropagation`

(bug fix)- tree based models where
`min_weight_fraction_leaf`

is used (enhancement) `model_selection.StratifiedKFold`

with`shuffle=True`

(this change, due to #7823 was not mentioned in the release notes at the time)

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¶

### New features¶

Classifiers and regressors

- Added
`multioutput.ClassifierChain`

for multi-label classification. By Adam Kleczewski. - Added solver
`'saga'`

that implements the improved version of Stochastic Average Gradient, in`linear_model.LogisticRegression`

and`linear_model.Ridge`

. It allows the use of L1 penalty with multinomial logistic loss, and behaves marginally better than ‘sag’ during the first epochs of ridge and logistic regression. #8446 by Arthur Mensch.

Other estimators

- Added the
`neighbors.LocalOutlierFactor`

class for anomaly detection based on nearest neighbors. #5279 by Nicolas Goix and Alexandre Gramfort. - Added
`preprocessing.QuantileTransformer`

class and`preprocessing.quantile_transform`

function for features normalization based on quantiles. #8363 by Denis Engemann, Guillaume Lemaitre, Olivier Grisel, Raghav RV, Thierry Guillemot, and Gael Varoquaux. - The new solver
`'mu'`

implements a Multiplicate Update in`decomposition.NMF`

, allowing the optimization of all beta-divergences, including the Frobenius norm, the generalized Kullback-Leibler divergence and the Itakura-Saito divergence. #5295 by Tom Dupre la Tour.

Model selection and evaluation

`model_selection.GridSearchCV`

and`model_selection.RandomizedSearchCV`

now support simultaneous evaluation of multiple metrics. Refer to the Specifying multiple metrics for evaluation section of the user guide for more information. #7388 by Raghav RV- Added the
`model_selection.cross_validate`

which allows evaluation of multiple metrics. This function returns a dict with more useful information from cross-validation such as the train scores, fit times and score times. Refer to The cross_validate function and multiple metric evaluation section of the userguide for more information. #7388 by Raghav RV - Added
`metrics.mean_squared_log_error`

, which computes the mean square error of the logarithmic transformation of targets, particularly useful for targets with an exponential trend. #7655 by Karan Desai. - Added
`metrics.dcg_score`

and`metrics.ndcg_score`

, which compute Discounted cumulative gain (DCG) and Normalized discounted cumulative gain (NDCG). #7739 by David Gasquez. - Added the
`model_selection.RepeatedKFold`

and`model_selection.RepeatedStratifiedKFold`

. #8120 by Neeraj Gangwar.

Miscellaneous

- Validation that input data contains no NaN or inf can now be suppressed
using
`config_context`

, at your own risk. This will save on runtime, and may be particularly useful for prediction time. #7548 by Joel Nothman. - Added a test to ensure parameter listing in docstrings match the function/class signature. #9206 by Alexandre Gramfort and Raghav RV.

### Enhancements¶

Trees and ensembles

- The
`min_weight_fraction_leaf`

constraint in tree construction is now more efficient, taking a fast path to declare a node a leaf if its weight is less than 2 * the minimum. Note that the constructed tree will be different from previous versions where`min_weight_fraction_leaf`

is used. #7441 by Nelson Liu. `ensemble.GradientBoostingClassifier`

and`ensemble.GradientBoostingRegressor`

now support sparse input for prediction. #6101 by Ibraim Ganiev.`ensemble.VotingClassifier`

now allows changing estimators by using`ensemble.VotingClassifier.set_params`

. An estimator can also be removed by setting it to`None`

. #7674 by Yichuan Liu.`tree.export_graphviz`

now shows configurable number of decimal places. #8698 by Guillaume Lemaitre.- Added
`flatten_transform`

parameter to`ensemble.VotingClassifier`

to change output shape of transform method to 2 dimensional. #7794 by Ibraim Ganiev and Herilalaina Rakotoarison.

Linear, kernelized and related models

`linear_model.SGDClassifier`

,`linear_model.SGDRegressor`

,`linear_model.PassiveAggressiveClassifier`

,`linear_model.PassiveAggressiveRegressor`

and`linear_model.Perceptron`

now expose`max_iter`

and`tol`

parameters, to handle convergence more precisely.`n_iter`

parameter is deprecated, and the fitted estimator exposes a`n_iter_`

attribute, with actual number of iterations before convergence. #5036 by Tom Dupre la Tour.- Added
`average`

parameter to perform weight averaging in`linear_model.PassiveAggressiveClassifier`

. #4939 by Andrea Esuli. `linear_model.RANSACRegressor`

no longer throws an error when calling`fit`

if no inliers are found in its first iteration. Furthermore, causes of skipped iterations are tracked in newly added attributes,`n_skips_*`

. #7914 by Michael Horrell.- In
`gaussian_process.GaussianProcessRegressor`

, method`predict`

is a lot faster with`return_std=True`

. #8591 by Hadrien Bertrand. - Added
`return_std`

to`predict`

method of`linear_model.ARDRegression`

and`linear_model.BayesianRidge`

. #7838 by Sergey Feldman. - Memory usage enhancements: Prevent cast from float32 to float64 in:
`linear_model.MultiTaskElasticNet`

;`linear_model.LogisticRegression`

when using newton-cg solver; and`linear_model.Ridge`

when using svd, sparse_cg, cholesky or lsqr solvers. #8835, #8061 by Joan Massich and Nicolas Cordier and Thierry Guillemot.

Other predictors

- Custom metrics for the
`neighbors`

binary trees now have fewer constraints: they must take two 1d-arrays and return a float. #6288 by Jake Vanderplas. `algorithm='auto`

in`neighbors`

estimators now chooses the most appropriate algorithm for all input types and metrics. #9145 by Herilalaina Rakotoarison and Reddy Chinthala.

Decomposition, manifold learning and clustering

`cluster.MiniBatchKMeans`

and`cluster.KMeans`

now use significantly less memory when assigning data points to their nearest cluster center. #7721 by Jon Crall.`decomposition.PCA`

,`decomposition.IncrementalPCA`

and`decomposition.TruncatedSVD`

now expose the singular values from the underlying SVD. They are stored in the attribute`singular_values_`

, like in`decomposition.IncrementalPCA`

. #7685 by Tommy Löfstedt`decomposition.NMF`

now faster when`beta_loss=0`

. #9277 by @hongkahjun.- Memory improvements for method
`barnes_hut`

in`manifold.TSNE`

#7089 by Thomas Moreau and Olivier Grisel. - Optimization schedule improvements for Barnes-Hut
`manifold.TSNE`

so the results are closer to the one from the reference implementation lvdmaaten/bhtsne by Thomas Moreau and Olivier Grisel. - Memory usage enhancements: Prevent cast from float32 to float64 in
`decomposition.PCA`

and`decomposition.randomized_svd_low_rank`

. #9067 by Raghav RV.

Preprocessing and feature selection

- Added
`norm_order`

parameter to`feature_selection.SelectFromModel`

to enable selection of the norm order when`coef_`

is more than 1D. #6181 by Antoine Wendlinger. - Added ability to use sparse matrices in
`feature_selection.f_regression`

with`center=True`

. #8065 by Daniel LeJeune. - Small performance improvement to n-gram creation in
`feature_extraction.text`

by binding methods for loops and special-casing unigrams. #7567 by Jaye Doepke - Relax assumption on the data for the
`kernel_approximation.SkewedChi2Sampler`

. Since the Skewed-Chi2 kernel is defined on the open interval \((-skewedness; +\infty)^d\), the transform function should not check whether`X < 0`

but whether`X < -self.skewedness`

. #7573 by Romain Brault. - Made default kernel parameters kernel-dependent in
`kernel_approximation.Nystroem`

. #5229 by Saurabh Bansod and Andreas Müller.

Model evaluation and meta-estimators

`pipeline.Pipeline`

is now able to cache transformers within a pipeline by using the`memory`

constructor parameter. #7990 by Guillaume Lemaitre.`pipeline.Pipeline`

steps can now be accessed as attributes of its`named_steps`

attribute. #8586 by Herilalaina Rakotoarison.- Added
`sample_weight`

parameter to`pipeline.Pipeline.score`

. #7723 by Mikhail Korobov. - Added ability to set
`n_jobs`

parameter to`pipeline.make_union`

. A`TypeError`

will be raised for any other kwargs. #8028 by Alexander Booth. `model_selection.GridSearchCV`

,`model_selection.RandomizedSearchCV`

and`model_selection.cross_val_score`

now allow estimators with callable kernels which were previously prohibited. #8005 by Andreas Müller .`model_selection.cross_val_predict`

now returns output of the correct shape for all values of the argument`method`

. #7863 by Aman Dalmia.- Added
`shuffle`

and`random_state`

parameters to shuffle training data before taking prefixes of it based on training sizes in`model_selection.learning_curve`

. #7506 by Narine Kokhlikyan. `model_selection.StratifiedShuffleSplit`

now works with multioutput multiclass (or multilabel) data. #9044 by Vlad Niculae.- Speed improvements to
`model_selection.StratifiedShuffleSplit`

. #5991 by Arthur Mensch and Joel Nothman. - Add
`shuffle`

parameter to`model_selection.train_test_split`

. #8845 by themrmax `multioutput.MultiOutputRegressor`

and`multioutput.MultiOutputClassifier`

now support online learning using`partial_fit`

. :issue: 8053 by Peng Yu.- Add
`max_train_size`

parameter to`model_selection.TimeSeriesSplit`

#8282 by Aman Dalmia. - More clustering metrics are now available through
`metrics.get_scorer`

and`scoring`

parameters. #8117 by Raghav RV. - A scorer based on
`metrics.explained_variance_score`

is also available. #9259 by Hanmin Qin.

Metrics

`metrics.matthews_corrcoef`

now support multiclass classification. #8094 by Jon Crall.- Add
`sample_weight`

parameter to`metrics.cohen_kappa_score`

. #8335 by Victor Poughon.

Miscellaneous

`utils.check_estimator`

now attempts to ensure that methods transform, predict, etc. do not set attributes on the estimator. #7533 by Ekaterina Krivich.- Added type checking to the
`accept_sparse`

parameter in`utils.validation`

methods. This parameter now accepts only boolean, string, or list/tuple of strings.`accept_sparse=None`

is deprecated and should be replaced by`accept_sparse=False`

. #7880 by Josh Karnofsky. - Make it possible to load a chunk of an svmlight formatted file by
passing a range of bytes to
`datasets.load_svmlight_file`

. #935 by Olivier Grisel. `dummy.DummyClassifier`

and`dummy.DummyRegressor`

now accept non-finite features. #8931 by @Attractadore.

### Bug fixes¶

Trees and ensembles

- Fixed a memory leak in trees when using trees with
`criterion='mae'`

. #8002 by Raghav RV. - Fixed a bug where
`ensemble.IsolationForest`

uses an an incorrect formula for the average path length #8549 by Peter Wang. - Fixed a bug where
`ensemble.AdaBoostClassifier`

throws`ZeroDivisionError`

while fitting data with single class labels. #7501 by Dominik Krzeminski. - Fixed a bug in
`ensemble.GradientBoostingClassifier`

and`ensemble.GradientBoostingRegressor`

where a float being compared to`0.0`

using`==`

caused a divide by zero error. #7970 by He Chen. - Fix a bug where
`ensemble.GradientBoostingClassifier`

and`ensemble.GradientBoostingRegressor`

ignored the`min_impurity_split`

parameter. #8006 by Sebastian Pölsterl. - Fixed
`oob_score`

in`ensemble.BaggingClassifier`

. #8936 by Michael Lewis - Fixed excessive memory usage in prediction for random forests estimators. #8672 by Mike Benfield.
- Fixed a bug where
`sample_weight`

as a list broke random forests in Python 2 #8068 by @xor. - Fixed a bug where
`ensemble.IsolationForest`

fails when`max_features`

is less than 1. #5732 by Ishank Gulati. - Fix a bug where gradient boosting with
`loss='quantile'`

computed negative errors for negative values of`ytrue - ypred`

leading to wrong values when calling`__call__`

. #8087 by Alexis Mignon - Fix a bug where
`ensemble.VotingClassifier`

raises an error when a numpy array is passed in for weights. #7983 by Vincent Pham. - Fixed a bug where
`tree.export_graphviz`

raised an error when the length of features_names does not match n_features in the decision tree. #8512 by Li Li.

Linear, kernelized and related models

- Fixed a bug where
`linear_model.RANSACRegressor.fit`

may run until`max_iter`

if it finds a large inlier group early. #8251 by @aivision2020. - Fixed a bug where
`naive_bayes.MultinomialNB`

and`naive_bayes.BernoulliNB`

failed when`alpha=0`

. #5814 by Yichuan Liu and Herilalaina Rakotoarison. - Fixed a bug where
`linear_model.LassoLars`

does not give the same result as the LassoLars implementation available in R (lars library). #7849 by Jair Montoya Martinez. - Fixed a bug in
`linear_model.RandomizedLasso`

,`linear_model.Lars`

,`linear_model.LassoLars`

,`linear_model.LarsCV`

and`linear_model.LassoLarsCV`

, where the parameter`precompute`

was not used consistently across classes, and some values proposed in the docstring could raise errors. #5359 by Tom Dupre la Tour. - Fix inconsistent results between
`linear_model.RidgeCV`

and`linear_model.Ridge`

when using`normalize=True`

. #9302 by Alexandre Gramfort. - Fix a bug where
`linear_model.LassoLars.fit`

sometimes left`coef_`

as a list, rather than an ndarray. #8160 by CJ Carey. - Fix
`linear_model.BayesianRidge.fit`

to return ridge parameter`alpha_`

and`lambda_`

consistent with calculated coefficients`coef_`

and`intercept_`

. #8224 by Peter Gedeck. - Fixed a bug in
`svm.OneClassSVM`

where it returned floats instead of integer classes. #8676 by Vathsala Achar. - Fix AIC/BIC criterion computation in
`linear_model.LassoLarsIC`

. #9022 by Alexandre Gramfort and Mehmet Basbug. - Fixed a memory leak in our LibLinear implementation. #9024 by Sergei Lebedev
- Fix bug where stratified CV splitters did not work with
`linear_model.LassoCV`

. #8973 by Paulo Haddad. - Fixed a bug in
`gaussian_process.GaussianProcessRegressor`

when the standard deviation and covariance predicted without fit would fail with a unmeaningful error by default. #6573 by Quazi Marufur Rahman and Manoj Kumar.

Other predictors

- Fix
`semi_supervised.BaseLabelPropagation`

to correctly implement`LabelPropagation`

and`LabelSpreading`

as done in the referenced papers. #9239 by Andre Ambrosio Boechat, Utkarsh Upadhyay, and Joel Nothman.

Decomposition, manifold learning and clustering

- Fixed the implementation of
`manifold.TSNE`

: `early_exageration`

parameter had no effect and is now used for the first 250 optimization iterations.- Fixed the
`AssertionError: Tree consistency failed`

exception reported in #8992. - Improve the learning schedule to match the one from the reference implementation lvdmaaten/bhtsne. by Thomas Moreau and Olivier Grisel.
- Fix a bug in
`decomposition.LatentDirichletAllocation`

where the`perplexity`

method was returning incorrect results because the`transform`

method returns normalized document topic distributions as of version 0.18. #7954 by Gary Foreman. - Fix output shape and bugs with n_jobs > 1 in
`decomposition.SparseCoder`

transform and`decomposition.sparse_encode`

for one-dimensional data and one component. This also impacts the output shape of`decomposition.DictionaryLearning`

. #8086 by Andreas Müller. - Fixed the implementation of
`explained_variance_`

in`decomposition.PCA`

,`decomposition.RandomizedPCA`

and`decomposition.IncrementalPCA`

. #9105 by Hanmin Qin. - Fixed the implementation of
`noise_variance_`

in`decomposition.PCA`

. #9108 by Hanmin Qin. - Fixed a bug where
`cluster.DBSCAN`

gives incorrect result when input is a precomputed sparse matrix with initial rows all zero. #8306 by Akshay Gupta - Fix a bug regarding fitting
`cluster.KMeans`

with a sparse array X and initial centroids, where X’s means were unnecessarily being subtracted from the centroids. #7872 by Josh Karnofsky. - Fixes to the input validation in
`covariance.EllipticEnvelope`

. #8086 by Andreas Müller. - Fixed a bug in
`covariance.MinCovDet`

where inputting data that produced a singular covariance matrix would cause the helper method`_c_step`

to throw an exception. #3367 by Jeremy Steward - Fixed a bug in
`manifold.TSNE`

affecting convergence of the gradient descent. #8768 by David DeTomaso. - Fixed a bug in
`manifold.TSNE`

where it stored the incorrect`kl_divergence_`

. #6507 by Sebastian Saeger. - Fixed improper scaling in
`cross_decomposition.PLSRegression`

with`scale=True`

. #7819 by jayzed82. `cluster.bicluster.SpectralCoclustering`

and`cluster.bicluster.SpectralBiclustering`

`fit`

method conforms with API by accepting`y`

and returning the object. #6126, #7814 by Laurent Direr and Maniteja Nandana.- Fix bug where
`mixture`

`sample`

methods did not return as many samples as requested. #7702 by Levi John Wolf. - Fixed the shrinkage implementation in
`neighbors.NearestCentroid`

. #9219 by Hanmin Qin.

Preprocessing and feature selection

- For sparse matrices,
`preprocessing.normalize`

with`return_norm=True`

will now raise a`NotImplementedError`

with ‘l1’ or ‘l2’ norm and with norm ‘max’ the norms returned will be the same as for dense matrices. #7771 by Ang Lu. - Fix a bug where
`feature_selection.SelectFdr`

did not exactly implement Benjamini-Hochberg procedure. It formerly may have selected fewer features than it should. #7490 by Peng Meng. - Fixed a bug where
`linear_model.RandomizedLasso`

and`linear_model.RandomizedLogisticRegression`

breaks for sparse input. #8259 by Aman Dalmia. - Fix a bug where
`feature_extraction.FeatureHasher`

mandatorily applied a sparse random projection to the hashed features, preventing the use of`feature_extraction.text.HashingVectorizer`

in a pipeline with`feature_extraction.text.TfidfTransformer`

. #7565 by Roman Yurchak. - Fix a bug where
`feature_selection.mutual_info_regression`

did not correctly use`n_neighbors`

. #8181 by Guillaume Lemaitre.

Model evaluation and meta-estimators

- Fixed a bug where
`model_selection.BaseSearchCV.inverse_transform`

returns`self.best_estimator_.transform()`

instead of`self.best_estimator_.inverse_transform()`

. #8344 by Akshay Gupta and Rasmus Eriksson. - Added
`classes_`

attribute to`model_selection.GridSearchCV`

,`model_selection.RandomizedSearchCV`

,`grid_search.GridSearchCV`

, and`grid_search.RandomizedSearchCV`

that matches the`classes_`

attribute of`best_estimator_`

. #7661 and #8295 by Alyssa Batula, Dylan Werner-Meier, and Stephen Hoover. - Fixed a bug where
`model_selection.validation_curve`

reused the same estimator for each parameter value. #7365 by Aleksandr Sandrovskii. `model_selection.permutation_test_score`

now works with Pandas types. #5697 by Stijn Tonk.- Several fixes to input validation in
`multiclass.OutputCodeClassifier`

#8086 by Andreas Müller. `multiclass.OneVsOneClassifier`

’s`partial_fit`

now ensures all classes are provided up-front. #6250 by Asish Panda.- Fix
`multioutput.MultiOutputClassifier.predict_proba`

to return a list of 2d arrays, rather than a 3d array. In the case where different target columns had different numbers of classes, a`ValueError`

would be raised on trying to stack matrices with different dimensions. #8093 by Peter Bull. - Cross validation now works with Pandas datatypes that that have a read-only index. #9507 by Loic Esteve.

Metrics

`metrics.average_precision_score`

no longer linearly interpolates between operating points, and instead weighs precisions by the change in recall since the last operating point, as per the Wikipedia entry. (#7356). By Nick Dingwall and Gael Varoquaux.- Fix a bug in
`metrics.classification._check_targets`

which would return`'binary'`

if`y_true`

and`y_pred`

were both`'binary'`

but the union of`y_true`

and`y_pred`

was`'multiclass'`

. #8377 by Loic Esteve. - Fixed an integer overflow bug in
`metrics.confusion_matrix`

and hence`metrics.cohen_kappa_score`

. #8354, #7929 by Joel Nothman and Jon Crall. - Fixed passing of
`gamma`

parameter to the`chi2`

kernel in`metrics.pairwise.pairwise_kernels`

#5211 by Nick Rhinehart, Saurabh Bansod and Andreas Müller.

Miscellaneous

- Fixed a bug when
`datasets.make_classification`

fails when generating more than 30 features. #8159 by Herilalaina Rakotoarison. - Fixed a bug where
`datasets.make_moons`

gives an incorrect result when`n_samples`

is odd. #8198 by Josh Levy. - Some
`fetch_`

functions in`datasets`

were ignoring the`download_if_missing`

keyword. #7944 by Ralf Gommers. - Fix estimators to accept a
`sample_weight`

parameter of type`pandas.Series`

in their`fit`

function. #7825 by Kathleen Chen. - Fix a bug in cases where
`numpy.cumsum`

may be numerically unstable, raising an exception if instability is identified. #7376 and #7331 by Joel Nothman and @yangarbiter. - Fix a bug where
`base.BaseEstimator.__getstate__`

obstructed pickling customizations of child-classes, when used in a multiple inheritance context. #8316 by Holger Peters. - Update Sphinx-Gallery from 0.1.4 to 0.1.7 for resolving links in documentation build with Sphinx>1.5 #8010, #7986 by Oscar Najera
- Add
`data_home`

parameter to`sklearn.datasets.fetch_kddcup99`

. #9289 by Loic Esteve. - Fix dataset loaders using Python 3 version of makedirs to also work in Python 2. #9284 by Sebastin Santy.
- Several minor issues were fixed with thanks to the alerts of [lgtm.com](http://lgtm.com). #9278 by Jean Helie, among others.

## API changes summary¶

Trees and ensembles

- Gradient boosting base models are no longer estimators. By Andreas Müller.
- All tree based estimators now accept a
`min_impurity_decrease`

parameter in lieu of the`min_impurity_split`

, which is now deprecated. The`min_impurity_decrease`

helps stop splitting the nodes in which the weighted impurity decrease from splitting is no longer at least`min_impurity_decrease`

. #8449 by Raghav RV.

Linear, kernelized and related models

`n_iter`

parameter is deprecated in`linear_model.SGDClassifier`

,`linear_model.SGDRegressor`

,`linear_model.PassiveAggressiveClassifier`

,`linear_model.PassiveAggressiveRegressor`

and`linear_model.Perceptron`

. By Tom Dupre la Tour.

Other predictors

`neighbors.LSHForest`

has been deprecated and will be removed in 0.21 due to poor performance. #9078 by Laurent Direr.`neighbors.NearestCentroid`

no longer purports to support`metric='precomputed'`

which now raises an error. #8515 by Sergul Aydore.- The
`alpha`

parameter of`semi_supervised.LabelPropagation`

now has no effect and is deprecated to be removed in 0.21. #9239 by Andre Ambrosio Boechat, Utkarsh Upadhyay, and Joel Nothman.

Decomposition, manifold learning and clustering

- Deprecate the
`doc_topic_distr`

argument of the`perplexity`

method in`decomposition.LatentDirichletAllocation`

because the user no longer has access to the unnormalized document topic distribution needed for the perplexity calculation. #7954 by Gary Foreman. - The
`n_topics`

parameter of`decomposition.LatentDirichletAllocation`

has been renamed to`n_components`

and will be removed in version 0.21. #8922 by @Attractadore. `decomposition.SparsePCA.transform`

’s`ridge_alpha`

parameter is deprecated in preference for class parameter. #8137 by Naoya Kanai.`cluster.DBSCAN`

now has a`metric_params`

parameter. #8139 by Naoya Kanai.

Preprocessing and feature selection

`feature_selection.SelectFromModel`

now has a`partial_fit`

method only if the underlying estimator does. By Andreas Müller.`feature_selection.SelectFromModel`

now validates the`threshold`

parameter and sets the`threshold_`

attribute during the call to`fit`

, and no longer during the call to`transform``

. By Andreas Müller.- The
`non_negative`

parameter in`feature_extraction.FeatureHasher`

has been deprecated, and replaced with a more principled alternative,`alternate_sign`

. #7565 by Roman Yurchak. `linear_model.RandomizedLogisticRegression`

, and`linear_model.RandomizedLasso`

have been deprecated and will be removed in version 0.21. #8995 by Ramana.S.

Model evaluation and meta-estimators

- Deprecate the
`fit_params`

constructor input to the`model_selection.GridSearchCV`

and`model_selection.RandomizedSearchCV`

in favor of passing keyword parameters to the`fit`

methods of those classes. Data-dependent parameters needed for model training should be passed as keyword arguments to`fit`

, and conforming to this convention will allow the hyperparameter selection classes to be used with tools such as`model_selection.cross_val_predict`

. #2879 by Stephen Hoover. - In version 0.21, the default behavior of splitters that use the
`test_size`

and`train_size`

parameter will change, such that specifying`train_size`

alone will cause`test_size`

to be the remainder. #7459 by Nelson Liu. `multiclass.OneVsRestClassifier`

now has`partial_fit`

,`decision_function`

and`predict_proba`

methods only when the underlying estimator does. #7812 by Andreas Müller and Mikhail Korobov.`multiclass.OneVsRestClassifier`

now has a`partial_fit`

method only if the underlying estimator does. By Andreas Müller.- The
`decision_function`

output shape for binary classification in`multiclass.OneVsRestClassifier`

and`multiclass.OneVsOneClassifier`

is now`(n_samples,)`

to conform to scikit-learn conventions. #9100 by Andreas Müller. - The
`multioutput.MultiOutputClassifier.predict_proba`

function used to return a 3d array (`n_samples`

,`n_classes`

,`n_outputs`

). In the case where different target columns had different numbers of classes, a`ValueError`

would be raised on trying to stack matrices with different dimensions. This function now returns a list of arrays where the length of the list is`n_outputs`

, and each array is (`n_samples`

,`n_classes`

) for that particular output. #8093 by Peter Bull. - Replace attribute
`named_steps`

`dict`

to`utils.Bunch`

in`pipeline.Pipeline`

to enable tab completion in interactive environment. In the case conflict value on`named_steps`

and`dict`

attribute,`dict`

behavior will be prioritized. #8481 by Herilalaina Rakotoarison.

Miscellaneous

Deprecate the

`y`

parameter in`transform`

and`inverse_transform`

. The method should not accept`y`

parameter, as it’s used at the prediction time. #8174 by Tahar Zanouda, Alexandre Gramfort and Raghav RV.SciPy >= 0.13.3 and NumPy >= 1.8.2 are now the minimum supported versions for scikit-learn. The following backported functions in

`utils`

have been removed or deprecated accordingly. #8854 and #8874 by Naoya KanaiThe

`store_covariances`

and`covariances_`

parameters of`discriminant_analysis.QuadraticDiscriminantAnalysis`

has been renamed to`store_covariance`

and`covariance_`

to be consistent with the corresponding parameter names of the`discriminant_analysis.LinearDiscriminantAnalysis`

. They will be removed in version 0.21. #7998 by JiachengRemoved in 0.19:

`utils.fixes.argpartition`

`utils.fixes.array_equal`

`utils.fixes.astype`

`utils.fixes.bincount`

`utils.fixes.expit`

`utils.fixes.frombuffer_empty`

`utils.fixes.in1d`

`utils.fixes.norm`

`utils.fixes.rankdata`

`utils.fixes.safe_copy`

Deprecated in 0.19, to be removed in 0.21:

`utils.arpack.eigs`

`utils.arpack.eigsh`

`utils.arpack.svds`

`utils.extmath.fast_dot`

`utils.extmath.logsumexp`

`utils.extmath.norm`

`utils.extmath.pinvh`

`utils.graph.graph_laplacian`

`utils.random.choice`

`utils.sparsetools.connected_components`

`utils.stats.rankdata`

Estimators with both methods

`decision_function`

and`predict_proba`

are now required to have a monotonic relation between them. The method`check_decision_proba_consistency`

has been added in**utils.estimator_checks**to check their consistency. #7578 by Shubham BhardwajAll checks in

`utils.estimator_checks`

, in particular`utils.estimator_checks.check_estimator`

now accept estimator instances. Most other checks do not accept estimator classes any more. #9019 by Andreas Müller.Ensure that estimators’ attributes ending with

`_`

are not set in the constructor but only in the`fit`

method. Most notably, ensemble estimators (deriving from`ensemble.BaseEnsemble`

) now only have`self.estimators_`

available after`fit`

. #7464 by Lars Buitinck and Loic Esteve.

## Code and Documentation Contributors¶

Thanks to everyone who has contributed to the maintenance and improvement of the project since version 0.18, including:

Joel Nothman, Loic Esteve, Andreas Mueller, Guillaume Lemaitre, Olivier Grisel, Hanmin Qin, Raghav RV, Alexandre Gramfort, themrmax, Aman Dalmia, Gael Varoquaux, Naoya Kanai, Tom Dupré la Tour, Rishikesh, Nelson Liu, Taehoon Lee, Nelle Varoquaux, Aashil, Mikhail Korobov, Sebastin Santy, Joan Massich, Roman Yurchak, RAKOTOARISON Herilalaina, Thierry Guillemot, Alexandre Abadie, Carol Willing, Balakumaran Manoharan, Josh Karnofsky, Vlad Niculae, Utkarsh Upadhyay, Dmitry Petrov, Minghui Liu, Srivatsan, Vincent Pham, Albert Thomas, Jake VanderPlas, Attractadore, JC Liu, alexandercbooth, chkoar, Óscar Nájera, Aarshay Jain, Kyle Gilliam, Ramana Subramanyam, CJ Carey, Clement Joudet, David Robles, He Chen, Joris Van den Bossche, Karan Desai, Katie Luangkote, Leland McInnes, Maniteja Nandana, Michele Lacchia, Sergei Lebedev, Shubham Bhardwaj, akshay0724, omtcyfz, rickiepark, waterponey, Vathsala Achar, jbDelafosse, Ralf Gommers, Ekaterina Krivich, Vivek Kumar, Ishank Gulati, Dave Elliott, ldirer, Reiichiro Nakano, Levi John Wolf, Mathieu Blondel, Sid Kapur, Dougal J. Sutherland, midinas, mikebenfield, Sourav Singh, Aseem Bansal, Ibraim Ganiev, Stephen Hoover, AishwaryaRK, Steven C. Howell, Gary Foreman, Neeraj Gangwar, Tahar, Jon Crall, dokato, Kathy Chen, ferria, Thomas Moreau, Charlie Brummitt, Nicolas Goix, Adam Kleczewski, Sam Shleifer, Nikita Singh, Basil Beirouti, Giorgio Patrini, Manoj Kumar, Rafael Possas, James Bourbeau, James A. Bednar, Janine Harper, Jaye, Jean Helie, Jeremy Steward, Artsiom, John Wei, Jonathan LIgo, Jonathan Rahn, seanpwilliams, Arthur Mensch, Josh Levy, Julian Kuhlmann, Julien Aubert, Jörn Hees, Kai, shivamgargsya, Kat Hempstalk, Kaushik Lakshmikanth, Kennedy, Kenneth Lyons, Kenneth Myers, Kevin Yap, Kirill Bobyrev, Konstantin Podshumok, Arthur Imbert, Lee Murray, toastedcornflakes, Lera, Li Li, Arthur Douillard, Mainak Jas, tobycheese, Manraj Singh, Manvendra Singh, Marc Meketon, MarcoFalke, Matthew Brett, Matthias Gilch, Mehul Ahuja, Melanie Goetz, Meng, Peng, Michael Dezube, Michal Baumgartner, vibrantabhi19, Artem Golubin, Milen Paskov, Antonin Carette, Morikko, MrMjauh, NALEPA Emmanuel, Namiya, Antoine Wendlinger, Narine Kokhlikyan, NarineK, Nate Guerin, Angus Williams, Ang Lu, Nicole Vavrova, Nitish Pandey, Okhlopkov Daniil Olegovich, Andy Craze, Om Prakash, Parminder Singh, Patrick Carlson, Patrick Pei, Paul Ganssle, Paulo Haddad, Paweł Lorek, Peng Yu, Pete Bachant, Peter Bull, Peter Csizsek, Peter Wang, Pieter Arthur de Jong, Ping-Yao, Chang, Preston Parry, Puneet Mathur, Quentin Hibon, Andrew Smith, Andrew Jackson, 1kastner, Rameshwar Bhaskaran, Rebecca Bilbro, Remi Rampin, Andrea Esuli, Rob Hall, Robert Bradshaw, Romain Brault, Aman Pratik, Ruifeng Zheng, Russell Smith, Sachin Agarwal, Sailesh Choyal, Samson Tan, Samuël Weber, Sarah Brown, Sebastian Pölsterl, Sebastian Raschka, Sebastian Saeger, Alyssa Batula, Abhyuday Pratap Singh, Sergey Feldman, Sergul Aydore, Sharan Yalburgi, willduan, Siddharth Gupta, Sri Krishna, Almer, Stijn Tonk, Allen Riddell, Theofilos Papapanagiotou, Alison, Alexis Mignon, Tommy Boucher, Tommy Löfstedt, Toshihiro Kamishima, Tyler Folkman, Tyler Lanigan, Alexander Junge, Varun Shenoy, Victor Poughon, Vilhelm von Ehrenheim, Aleksandr Sandrovskii, Alan Yee, Vlasios Vasileiou, Warut Vijitbenjaronk, Yang Zhang, Yaroslav Halchenko, Yichuan Liu, Yuichi Fujikawa, affanv14, aivision2020, xor, andreh7, brady salz, campustrampus, Agamemnon Krasoulis, ditenberg, elena-sharova, filipj8, fukatani, gedeck, guiniol, guoci, hakaa1, hongkahjun, i-am-xhy, jakirkham, jaroslaw-weber, jayzed82, jeroko, jmontoyam, jonathan.striebel, josephsalmon, jschendel, leereeves, martin-hahn, mathurinm, mehak-sachdeva, mlewis1729, mlliou112, mthorrell, ndingwall, nuffe, yangarbiter, plagree, pldtc325, Breno Freitas, Brett Olsen, Brian A. Alfano, Brian Burns, polmauri, Brandon Carter, Charlton Austin, Chayant T15h, Chinmaya Pancholi, Christian Danielsen, Chung Yen, Chyi-Kwei Yau, pravarmahajan, DOHMATOB Elvis, Daniel LeJeune, Daniel Hnyk, Darius Morawiec, David DeTomaso, David Gasquez, David Haberthür, David Heryanto, David Kirkby, David Nicholson, rashchedrin, Deborah Gertrude Digges, Denis Engemann, Devansh D, Dickson, Bob Baxley, Don86, E. Lynch-Klarup, Ed Rogers, Elizabeth Ferriss, Ellen-Co2, Fabian Egli, Fang-Chieh Chou, Bing Tian Dai, Greg Stupp, Grzegorz Szpak, Bertrand Thirion, Hadrien Bertrand, Harizo Rajaona, zxcvbnius, Henry Lin, Holger Peters, Icyblade Dai, Igor Andriushchenko, Ilya, Isaac Laughlin, Iván Vallés, Aurélien Bellet, JPFrancoia, Jacob Schreiber, Asish Mahapatra

# Previous Releases¶

- Version 0.18.2
- Version 0.18.1
- Version 0.18
- Version 0.17.1
- Version 0.17
- Version 0.16.1
- Version 0.16
- Version 0.15.2
- Version 0.15.1
- Version 0.15
- Version 0.14
- Version 0.13.1
- Version 0.13
- Version 0.12.1
- Version 0.12
- Version 0.11
- Version 0.10
- Version 0.9
- Version 0.8
- Version 0.7
- Version 0.6
- Version 0.5
- Version 0.4
- Earlier versions