API reference¶

Auto-generated from Python docstrings. Pages are organized by family rather than alphabetically — most users want “all the logistic estimators” or “all the LS estimators with groups”, not a linear scan through 80 classes.

Estimators¶

Family	Page
Gaussian (LS) — scalar penalties (MCP, SCAD, ElasticNet)	LS
Gaussian (LS) — group penalties (incl. GroupElasticNet)	LS
Multi-task LS — 2D `Y`, joint feature selection	Multi-task
Adaptive {Lasso, MCP, SCAD} — pilot-fit two-stage	Adaptive
Binomial logistic	Logistic
Multinomial / softmax (K classes, row-grouped)	Multinomial
Poisson	Poisson
Cox proportional hazards	Cox
Graphical models — sparse precision matrices	Graphical

Every GLM family follows the same pattern: scalar (MCP, SCAD) and group (GroupLasso, GroupMCP, SparseGroupLasso, SparseGroupMCP) penalties; each penalty has a single-λ Regressor and a full-path PathRegressor. The Gaussian LS family also adds ElasticNet (scalar) and GroupElasticNet (group). The multi-task LS family adds 8 more (MultiTaskLasso/MCP/SCAD/ElasticNet × {single-λ, Path}).

Cross-validation, IC selection, and stability selection¶

Module	Page
`*PathCV` cross-validation (every CV class accepts `n_jobs` for threaded folds)	CV
`select_by_ic` (AIC/BIC/EBIC)	IC
`StabilitySelection` (MB bootstrap)	Stability
`ebic_path`, `joint_ebic_path` (graphical models)	Graph selection
`GraphicalStabilitySelection`, `GraphicalBootstrap` (bootnet-style edge bootstrap)	Graph stability

Inference¶

Module	Page
`debiased_lasso`, `debiased_logistic_lasso`, `debiased_poisson_lasso` (Wald CIs / p-values via VBR nodewise)	Debiased

Design-matrix helpers¶

Module	Page
`MmapDesignF64`, `MmapDesignF32`, `ChunkedDesignF64`, `ChunkedDesignF32`	Design

Extension ABCs¶

Module	Page
`skein_glm.penalties.Penalty`, `skein_glm.penalties.GroupPenalty`	ABCs
`skein_glm.datafits.Datafit`	ABCs

A note on inherited members¶

Estimator subclasses share most of their behavior through base classes (_PathRegressorBase, _LogisticPathRegressorBase, etc.). The auto-generated docs show each concrete class with a brief docstring; for the meatier “what does fit actually do” content, look at the base class section near the top of each family page.

Some methods (sklearn’s BaseEstimator.get_params, set_params, __sklearn_tags__) are inherited from sklearn itself. We don’t re-document those — see sklearn’s docs for the inheritance chain.