Ensure all attributes are documented

[amueller]

as discussed in #13385 we need to ensure all attributes are documented.

if you want to work on this, you should pick a specific submodule and fix all the attribute documentation mismatches in that submodule.

Here's a script to find remaining ones (there might be some false positives):

import numpy as np
from sklearn.base import clone
from sklearn.utils.testing import all_estimators
from sklearn.utils.estimator_checks import pairwise_estimator_convert_X, enforce_estimator_tags_y
from numpydoc import docscrape

ests = all_estimators()

for name, Est in ests:
    try:
        estimator_orig = Est()
    except:
        continue
    rng = np.random.RandomState(0)
    X = pairwise_estimator_convert_X(rng.rand(40, 10), estimator_orig)
    X = X.astype(object)
    y = (X[:, 0] * 4).astype(np.int)
    est = clone(estimator_orig)
    y = enforce_estimator_tags_y(est, y)
    try:
        est.fit(X, y)
    except:
        continue
    fitted_attrs = [(x, getattr(est, x, None))
                    for x in est.__dict__.keys() if x.endswith("_")
                    and not x.startswith("_")]
    doc = docscrape.ClassDoc(type(est))
    doc_attributes = []
    incorrect = []
    for att_name, type_definition, param_doc in doc['Attributes']:
        if not type_definition.strip():
            if ':' in att_name and att_name[:att_name.index(':')][-1:].strip():
                incorrect += [name +
                              ' There was no space between the param name and '
                              'colon (%r)' % att_name]
            elif name.rstrip().endswith(':'):
                incorrect += [name +
                              ' Parameter %r has an empty type spec. '
                              'Remove the colon' % (att_name.lstrip())]

        if '*' not in att_name:
            doc_attributes.append(att_name.split(':')[0].strip('` '))
    assert incorrect == []
    fitted_attrs_names = [x[0] for x in fitted_attrs]

    bad = sorted(list(set(fitted_attrs_names) ^ set(doc_attributes)))
    if len(bad) > 0:
        msg = '{}\n'.format(name) + '\n'.join(bad)
        print("Docstring Error: Attribute mismatch in " + msg)

[alexitkes]

I have already found at least one mismatch in attribute documentation in NMF class description. I think I can take some of this work. I am almost ready to propose some changes within decomposition and random_projection submodules.

[thomasjpfan]

Missing attribute docstrings for each estimator

Reference this issue in your PR

• ARDRegression, [intercept_]
• AdaBoostClassifier, [base_estimator_]
• AdaBoostRegressor, [base_estimator_]
• AdditiveChi2Sampler, [sample_interval_]
• AgglomerativeClustering, [n_components_] (deprecated)
• BaggingClassifier, [n_features_]
• BaggingRegressor, [base_estimator_, n_features_]
• BayesianGaussianMixture, [mean_precision_prior, mean_precision_prior_]
• BayesianRidge, [X_offset_, X_scale_]
• BernoulliNB, [coef_, intercept_]
• BernoulliRBM, [h_samples_]
• Birch, [fit_, partial_fit_]
• CCA, [coef_, x_mean_, x_std_, y_mean_, y_std_]
• CheckingClassifier, [classes_]
• ComplementNB, [coef_, intercept_]
• CountVectorizer, [stop_words_, vocabulary_]
• DecisionTreeRegressor, [classes_, n_classes_]
• DictVectorizer, [feature_names_, vocabulary_]
• DummyClassifier, [output_2d_]
• DummyRegressor, [output_2d_]
• ElasticNet, [dual_gap_]
• ElasticNetCV, [dual_gap_]
• EllipticEnvelope, [dist_, raw_covariance_, raw_location_, raw_support_]
• ExtraTreeClassifier, [feature_importances_]
• ExtraTreeRegressor, [classes_, feature_importances_, n_classes_]
• ExtraTreesClassifier, [base_estimator_]
• ExtraTreesRegressor, [base_estimator_]
• FactorAnalysis, [mean_]
• FeatureAgglomeration, [n_components_]
• GaussianProcessClassifier, [base_estimator_]
• GaussianRandomProjection, [components_]
• GradientBoostingClassifier, [max_features_, n_classes_, n_features_, oob_improvement_]
• GradientBoostingRegressor, [max_features_, n_classes_, n_estimators_, n_features_, oob_improvement_]
• HistGradientBoostingClassifier, [bin_mapper_, classes_, do_early_stopping_, loss_, n_features_, scorer_]
• HistGradientBoostingRegressor, [bin_mapper_, do_early_stopping_, loss_, n_features_, scorer_]
• IncrementalPCA, [batch_size_]
• IsolationForest, [base_estimator_, estimators_features_, n_features_]
• IsotonicRegression, [X_max_, X_min_, f_]
• IterativeImputer, [random_state_]
• KNeighborsClassifier, [classes_, effective_metric_, effective_metric_params_, outputs_2d_]
• KNeighborsRegressor, [effective_metric_, effective_metric_params_]
• KernelCenterer, [K_fit_all_, K_fit_rows_]
• KernelDensity, [tree_]
• KernelPCA, [X_transformed_fit_, dual_coef_]
• LabelBinarizer, [classes_, sparse_input_, y_type_]
• LabelEncoder, [classes_]
• LarsCV, [active_]
• Lasso, [dual_gap_]
• LassoLarsCV, [active_]
• LassoLarsIC, [alphas_]
• LatentDirichletAllocation, [bound_, doc_topic_prior_, exp_dirichlet_component_, random_state_, topic_word_prior_]
• LinearDiscriminantAnalysis, [covariance_]
• LinearRegression, [rank_, singular_]
• LinearSVC, [classes_]
• LocalOutlierFactor, [effective_metric_, effective_metric_params_]
• MDS, [dissimilarity_matrix_, n_iter_]
• MLPClassifier, [best_loss_, loss_curve_, t_]
• MLPRegressor, [best_loss_, loss_curve_, t_]
• MinMaxScaler, [n_samples_seen_]
• MiniBatchDictionaryLearning, [iter_offset_]
• MiniBatchKMeans, [counts_, init_size_, n_iter_]
• MultiLabelBinarizer, [classes_]
• MultiTaskElasticNet, [dual_gap_, eps_, sparse_coef_]
• MultiTaskElasticNetCV, [dual_gap_]
• MultiTaskLasso, [dual_gap_, eps_, sparse_coef_]
• MultiTaskLassoCV, [dual_gap_]
• NearestCentroid, [classes_]
• NearestNeighbors, [effective_metric_, effective_metric_params_]
• NeighborhoodComponentsAnalysis, [random_state_]
• NuSVC, [class_weight_, fit_status_, probA_, probB_, shape_fit_]
• NuSVR, [class_weight_, fit_status_, n_support_, probA_, probB_, shape_fit_]
• OAS, [location_]
• OneClassSVM, [class_weight_, fit_status_, n_support_, probA_, probB_, shape_fit_]
• OneVsOneClassifier, [n_classes_]
• OneVsRestClassifier, [coef_, intercept_, n_classes_]
• OrthogonalMatchingPursuit, [n_nonzero_coefs_]
• PLSCanonical, [coef_, x_mean_, x_std_, y_mean_, y_std_]
• PLSRegression, [x_mean_, x_std_, y_mean_, y_std_]
• PLSSVD, [x_mean_, x_std_, y_mean_, y_std_]
• PassiveAggressiveClassifier, [loss_function_, t_]
• PassiveAggressiveRegressor, [t_]
• Perceptron, [loss_function_]
• QuadraticDiscriminantAnalysis, [classes_, covariance_]
• RBFSampler, [random_offset_, random_weights_]
• RFE, [classes_]
• RFECV, [classes_]
• RadiusNeighborsClassifier, [classes_, effective_metric_, effective_metric_params_, outputs_2d_]
• RadiusNeighborsRegressor, [effective_metric_, effective_metric_params_]
• RandomForestClassifier, [oob_decision_function_, oob_score_]
• RandomForestRegressor, [oob_prediction_, oob_score_]
• RandomTreesEmbedding, [base_estimator_, feature_importances_, n_features_, n_outputs_, one_hot_encoder_]
• RidgeCV, [cv_values_]
• RidgeClassifier, [classes_]
• RidgeClassifierCV, [cv_values_]
• SGDClassifier, [classes_, t_]
• SGDRegressor, [average_coef_, average_intercept_]
• SVC, [class_weight_, shape_fit_]
• SVR, [class_weight_, fit_status_, n_support_, probA_, probB_, shape_fit_]
• SelectKBest, [pvalues_, scores_]
• ShrunkCovariance, [shrinkage]
• SkewedChi2Sampler, [random_offset_, random_weights_]
• SparseRandomProjection, [components_, density_]
• SpectralEmbedding, [n_neighbors_]
• TfidfVectorizer, [stop_words_, vocabulary_]

[mepa]

I can take up the tree submodule attribute documentation mismatches, which includes:

• DecisionTreeRegressor, [classes_, n_classes_]
• ExtraTreeClassifier, [classes_, max_features_, n_classes_, n_features_, n_outputs_, tree_]
• ExtraTreeRegressor, [classes_, max_features_, n_classes_, n_features_, n_outputs_, tree_]

[wendyhhu]

I'm working on LinearRegression, [rank_, singular_].

[wendyhhu]

I'm working on LinearSVC, [n_iter_] and LinearSVR, [n_iter_]

[matsmaiwald]

I'll take up Gradient boosting i.e.

• GradientBoostingClassifier [base_estimator_, max_features_, n_classes_, n_features_]
• GradientBoostingRegressor [base_estimator_, classes_, max_features_, n_estimators_, n_features_]

[matsmaiwald]

nevermind, misread where attributes are missing and where not

[alexitkes]

It's looking like there is also classes_ attribute undocumented for classifiers of naive_bayes submodule. I have started to fix it.

[mandalbiswadip]

I will work on TfidfVectorizer, [fixed_vocabulary_]

[rcwoolston]

I will work on:

• RandomForestClassifier, [base_estimator_]
• RandomForestRegressor, [base_estimator_, n_classes_]
• ExtraTreesClassifier, [base_estimator_]
• ExtraTreesRegressor, [base_estimator_, n_classes_]

[wendyhhu]

I'm working on:

• SGDClassifier, [average_coef_, average_intercept_, standard_coef_, standard_intercept_]
• SGDRegressor, [standard_coef_, standard_intercept_]

EDIT: opened an issue to change these attributes from public to private (reference: #14364)

[SwordKnight6216]

I am working on:
KernelCenterer, [K_fit_all_, K_fit_rows_]
MinMaxScaler, [n_samples_seen_]

[rcwoolston]

I will work on:

• RandomTreesEmbedding, [base_estimator_, classes_, feature_importances_, n_classes_, n_features_, n_outputs_, one_hot_encoder_]

[marenwestermann]

I'm working on Lasso.

[marenwestermann]

I'm now working on adding the attribute sparse_coef_ to MultiTaskElasticNet and MultiTaskLasso.

[marenwestermann]

I'm working on LarsCV.

[marenwestermann]

@thomasjpfan it is said in the classes SVR and OneClassSVM:
"The probA_ attribute is deprecated in version 0.23 and will be removed in version 0.25." and
"The probB_ attribute is deprecated in version 0.23 and will be removed in version 0.25."

Therefore, these attributes probably don't need documentation anymore, right?
Going from here, will these two attributes also be deprecated in the class NuSVR?

[marenwestermann]

The attributes classes_ and n_classes_ for ExtraTreeRegressor are false positives.

[thomasjpfan]

Therefore, these attributes probably don't need documentation anymore, right?
Going from here, will these two attributes also be deprecated in the class NuSVR?

Since we are deprecating them I would say we would not need document them.

The attributes classes_ and n_classes_ for ExtraTreeRegressor are false positives.

Yup those should be deprecated then removed if they are not already.

[Abilityguy]

The DecisionTreeRegressor class says:
"the n_classes_ attribute is to be deprecated from version 0.22 and will be removed in 0.24."
"the classes_ attribute is to be deprecated from version 0.22 and will be removed in 0.24."

So these attributes also don't need documentation right?

[cmarmo]

So these attributes also don't need documentation right?

Right @Abilityguy, thanks for pointing out that.

[mynkdsi1011]

I can see below mismatch in RidgeGCV :
Docstring Error: Attribute mismatch in RidgeGCV
alpha
best_score
coef_
dual_coef_
intercept_
n_features_in_

and in BaseRidgeCV:
Docstring Error: Attribute mismatch in BaseRidgeCV
alpha
best_score
coef_
intercept_
n_features_in_

Can I take it up? I am first timer and wants to contribute.

[srivathsa729]

@marenwestermann in the class FeatureAgglomeration, it is said that, in version 0.21, n_connected_components_ was added to replace n_components_, then n_components_ would be false positive right..?

[marenwestermann]

@srivathsa729 from my understanding yes. However, it would be good if one of the core developers could double check.

[disha4u]

I will take up ElasticNet

[marenwestermann]

Documentation of the attributes X_offset_ and X_scale_ for BayesianRidge has been added with #18607 .

[marenwestermann]

The attribute output_2d_ is deprecated in DummyClassifier and DummyRegressor (see #14933).

[marenwestermann]

I ran the script provided by @amueller at the top of this PR (the code needs to be slightly modified because things have moved around). I couldn't find any more attributes that need to be documented with the exception of n_features_in_ which I see has been introduced in #16112. This attribute is undocumented in I think all classes it was introduced to. Should it be documented?
ping @NicolasHug

[ShyamDesai]

Hello. I wanted to take this on as a first issue, but it seems that all attributes have already been documented?

[cmarmo]

Thanks @marenwestermann for checking! This is very helpful.
n_features_in_ documentation is now tracked in #19333.

[cmarmo]

It turns out that all detections from the script in the descriptions are false positives, I'm closing this one. Thanks to all the contributors for their helpful work!