We should be using Sphinx's versionadded and versionchanged directives to indicate when classes, functions, methods and parameters are added or modified in their semantics. We are not very good at ensuring this labelling is done.

Towards the upcoming release of v0.22, it would be useful if some contributors (or sprinters) scoured the change log and identified whether any added/changed parameters/classes needed a versionadded or versionchanged label, and to produce pull requests adding them when otherwise omitted.

I also note past omissions, for example KBinsDiscretizer and IterativeImputer do not mention their versionadded, while ColumnTransformer does. Recording for each estimator when it was first released would be helpful for users when looking at the documentation.

Suggested procedure

Check the changelog in doc/whats_new/ one what's a new file at a time.
**Make a single PR for an estimator and a specific version. Do not include multiple estimators and multiple versions at once (it makes it difficult to review).

Focus on Public API: if a method has been made private, check the tickbox and move to another.

.. versionadded:: 0.xx

should be included only for new estimators/parameters

.. versionchanged:: 0.xx
   `param_xxx` change from 'xx' to 'yy'

should be included when the default value of a parameter change.

Classes lacking label

The list may be incomplete.

• ARDRegression PRs put in version_added label for ARDRegression (and many others): Issue #15426 #15482 put in version_added label #15973 adding versionadded to a few modules #16015
• AdaBoostClassifier
• AdaBoostRegressor
• AdamOptimizer
• AdditiveChi2Sampler
• AffinityPropagation
• AgglomerationTransform
• BaggingRegressor
• BaseBagging
• BaseCrossValidator
• BaseDecisionTree
• BaseDiscreteNB
• BaseEnsemble
• BaseForest
• BaseGradientBoosting
• BaseHistGradientBoosting
• BaseLabelPropagation
• BaseLibSVM
• BaseLoss
• BaseMixture
• BaseNB
• BaseRandomProjection
• BaseSGD
• BaseSVC
• BaseSearchCV
• BaseShuffleSplit
• BaseSpectral
• BaseWeightBoosting
• BernoulliNB
• BernoulliRBM
• Binarizer
• BinaryCrossEntropy
• BinomialDeviance
• BinomialDeviance
• Birch
• Bunch
• CCA
• CalibratedClassifierCV
• CategoricalCrossEntropy
• CategoricalNB
• ChangedBehaviorWarning
• CheckingClassifier
• ClassificationLossFunction
• ClassifierChain
• ComplementNB
• ConvergenceWarning
• CountVectorizer
• DataConversionWarning
• DataDimensionalityWarning
• DictVectorizer
• DummyRegressor
• ElasticNet
• ElasticNetCV
• EllipticEnvelope
• EmpiricalCovariance
• ExponentialLoss
• ExponentialLoss
• FactorAnalysis
• FastICA
• FeatureHasher
• FeatureUnion
• FitFailedWarning
• ForestClassifier
• ForestRegressor
• GaussianNB
• GaussianRandomProjection
• GenericUnivariateSelect
• GraphicalLasso
• GraphicalLassoCV
• GridSearchCV
• GroupKFold
• HistGradientBoostingClassifier
• HistGradientBoostingRegressor
• HuberLossFunction
• IncrementalPCA
• IsotonicRegression
• IterativeImputer Open PR
• KBinsDiscretizer
• KFold
• KMeans
• KNeighborsClassifier
• KNeighborsRegressor
• KernelCenterer
• KernelDensity
• KernelRidge
• KeyValTuple
• KeyValTupleParam
• LabelBinarizer
• LabelEncoder
• LabelPropagation
• LabelSpreading
• Lars
• LarsCV
• Lasso
• LassoCV
• LassoLars
• LassoLarsCV
• LassoLarsIC
• LeastAbsoluteDeviation
• LeastAbsoluteError
• LeastSquares
• LeastSquaresError
• LeaveOneGroupOut
• LeaveOneOut
• LeavePGroupsOut
• LeavePOut
• LedoitWolf
• LinearClassifierMixin
• LinearModel
• LinearModelCV
• LinearRegression
• LinearSVC
• LinearSVR
• LocalOutlierFactor
• LocallyLinearEmbedding
• LossFunction
• MDS
• MinCovDet
• MiniBatchKMeans
• MissingIndicator Open PR
• Module_six_moves_urllib
• Module_six_moves_urllib_error
• Module_six_moves_urllib_parse
• Module_six_moves_urllib_request
• Module_six_moves_urllib_response
• Module_six_moves_urllib_robotparser
• MultiLabelBinarizer
• MultiOutputClassifier
• MultiOutputRegressor
• MultiTaskElasticNet
• MultiTaskElasticNetCV
• MultiTaskLasso
• MultiTaskLassoCV
• MultinomialDeviance
• MultinomialNB
• NearestCentroid
• NearestNeighbors
• NeighborhoodComponentsAnalysis
• NeighborsBase
• NoSampleWeightWrapper
• NonBLASDotWarning
• Normalizer
• NotFittedError
• NuSVR
• Nystroem
• OAS
• OPTICS
• OneClassSVM
• OneHotEncoder
• OneVsOneClassifier
• OneVsRestClassifier
• OrdinalEncoder
• OrthogonalMatchingPursuit
• OrthogonalMatchingPursuitCV
• OutputCodeClassifier
• PLSCanonical
• PLSRegression
• PLSSVD
• PatchExtractor
• Pipeline
• PowerTransformer
• PredefinedSplit
• QuantileLossFunction
• QuantileTransformer
• RBFSampler
• RFE
• RFECV
• RadiusNeighborsClassifier
• RadiusNeighborsRegressor
• RandomizedSearchCV
• RegressionLossFunction
• RegressorChain
• RepeatedKFold
• RepeatedStratifiedKFold
• RidgeCV
• RidgeClassifierCV
• SGDOptimizer
• SVR
• ScaledLogOddsEstimator
• SelectFdr
• SelectFpr
• SelectFwe
• SelectKBest
• SelectPercentile
• SelectorMixin
• ShrunkCovariance
• ShuffleSplit
• SkewedChi2Sampler
• SkipTestWarning
• SimpleImputer Open PR
• SparseCodingMixin
• SparseRandomProjection
• SpectralBiclustering
• SpectralClustering
• SpectralCoclustering
• SpectralEmbedding
• StratifiedKFold
• StratifiedShuffleSplit
• TfidfTransformer PRs put in version_added label for ARDRegression (and many others): Issue #15426 #15482 put in version_added label #15973 adding versionadded to a few modules #16015
• TfidfVectorizer PRs put in version_added label for ARDRegression (and many others): Issue #15426 #15482 put in version_added label #15973 adding versionadded to a few modules #16015
• TheilSenRegressor
• TimeSeriesSplit
• TransformedTargetRegressor
• TruncatedSVD
• UndefinedMetricWarning
• VarianceThreshold