sklearn-api

name	sklearn-api
description	Look up the public API surface of any scikit-learn module — names, signatures, and one-line summaries. Use whenever you need to know what's exported from `sklearn.X`, recall a class/function signature, find the right estimator/metric/utility for a task, or check whether something is part of the public API. The reference is generated from `doc/api_reference.py`, so it matches what appears on the official docs site.
allowed-tools	["Read","Bash(ls *)","Bash(grep *)"]

sklearn-api — scikit-learn public API reference

This skill is a flat, file-per-module index of scikit-learn's public API. Each file under references/ lists the names that the documentation site exposes for one module, with their signature and a one-line summary pulled from the docstring.

The references are the source of truth for "is X public?" If a name is not in the relevant references/sklearn.<module>.md, treat it as private — even if it's importable. The list is generated from doc/api_reference.py, which is what gets rendered on scikit-learn.org.

How to use

1. Identify the module the user is asking about (or that you need to recall).

2. Read references/sklearn.<module>.md. For the top-level sklearn namespace itself (e.g. set_config), read references/sklearn.md.

3. If you don't know which module owns a name, grep across the references:

   grep -l "^### \`<Name>\`" .claude/skills/sklearn-api/references/
   

4. The reference gives you the signature and a one-liner. For full parameter docs, examples, or attributes, read the actual source under sklearn/ — the reference is for discovery, not for replacing the docstring.

Available module references

Top-level:

• sklearn.md — config_context, get_config, set_config, show_versions

Core building blocks:

• sklearn.base.md — base classes and mixins (BaseEstimator, *Mixin, clone)
• sklearn.exceptions.md — warning / error classes
• sklearn.pipeline.md — Pipeline, FeatureUnion, make_pipeline
• sklearn.compose.md — ColumnTransformer, TransformedTargetRegressor
• sklearn.frozen.md — FrozenEstimator
• sklearn.experimental.md — opt-in flags for experimental features

Estimators by family:

• sklearn.linear_model.md — linear / logistic / GLM / robust regressors
• sklearn.svm.md — SVC, SVR, NuSVC, LinearSVC, OneClassSVM
• sklearn.tree.md — decision tree estimators + export_*
• sklearn.ensemble.md — random forests, gradient boosting, stacking, voting
• sklearn.neighbors.md — KNN, radius neighbors, KD/Ball tree, LOF
• sklearn.naive_bayes.md — Gaussian / Multinomial / Bernoulli / Categorical NB
• sklearn.discriminant_analysis.md — LDA, QDA
• sklearn.gaussian_process.md — GP regressor/classifier + kernels
• sklearn.neural_network.md — MLP, BernoulliRBM
• sklearn.semi_supervised.md — LabelPropagation, LabelSpreading, SelfTrainingClassifier
• sklearn.dummy.md — DummyClassifier, DummyRegressor
• sklearn.kernel_ridge.md — KernelRidge
• sklearn.isotonic.md — IsotonicRegression + helpers
• sklearn.calibration.md — CalibratedClassifierCV, CalibrationDisplay
• sklearn.multiclass.md / sklearn.multioutput.md — meta-estimators

Unsupervised:

• sklearn.cluster.md — KMeans, DBSCAN, HDBSCAN, hierarchical, spectral, etc.
• sklearn.mixture.md — GaussianMixture, BayesianGaussianMixture
• sklearn.decomposition.md — PCA, NMF, ICA, dictionary learning, LDA
• sklearn.cross_decomposition.md — PLS family, CCA
• sklearn.manifold.md — TSNE, UMAP-style (Isomap, LLE, MDS, SpectralEmbedding)
• sklearn.covariance.md — robust / shrinkage covariance estimators
• sklearn.random_projection.md — Gaussian / sparse random projections

Data prep & feature work:

• sklearn.preprocessing.md — scalers, encoders, polynomial, splines, target encoding
• sklearn.impute.md — SimpleImputer, IterativeImputer, KNNImputer, MissingIndicator
• sklearn.feature_extraction.md — dict/hash vectorizers + image.* + text.* submodules
• sklearn.feature_selection.md — filter / wrapper / model-based selection
• sklearn.kernel_approximation.md — Nystroem, RBFSampler, etc.

Evaluation & selection:

• sklearn.metrics.md — every scorer, classification/regression/clustering metric, *Display, pairwise.*
• sklearn.model_selection.md — splitters, search CV, cross_val_*, learning/validation curves
• sklearn.inspection.md — permutation_importance, partial dependence, *Display

Data:

• sklearn.datasets.md — load_*, fetch_*, make_*, dump_svmlight_file

Utilities:

• sklearn.utils.md — validation, parallel, sparse helpers, multiclass utils, Bunch, fixes, etc.

Tips for common questions

• "What estimators does sklearn ship for X?" — open the family file (sklearn.linear_model.md, sklearn.ensemble.md, ...) and scan the section headings; they group estimators by sub-family.
• "What scorer / metric should I use for X?" — sklearn.metrics.md is segmented into Classification / Regression / Clustering / Pairwise / Plotting sections.
• "Is this thing public?" — search the references. If it's not there, it's not part of the documented API.
• Submodule entries (e.g. image.PatchExtractor inside sklearn.feature_extraction.md, or pairwise.cosine_similarity inside sklearn.metrics.md) live under their parent module's reference file; the prefix tells you the actual import path.

Regenerating

The reference files are generated, not hand-written. After a public-API change in scikit-learn (anything that touches doc/api_reference.py, constructor signatures, or first-line docstrings), re-run:

pixi run -e dev python .claude/skills/sklearn-api/generate_references.py

The generator walks API_REFERENCE from doc/api_reference.py, resolves each name via importlib, formats inspect.signature(...), and writes one markdown file per module key into references/.