Probabilistic supervised regression

The skpro.regression module contains algorithms and composition tools for probabilistic supervised regression, i.e., tabular regression estimation with a probabilistic prediction mode.

This learning task is sometimes also known as conditional distribution predictions, or conditional density estimation, if predictive distributions are continuous.

All regressors in skpro can be listed using the skpro.registry.all_objects utility, using object_types="regressor_proba", optionally filtered by tags. Valid tags can be listed using skpro.registry.all_tags.

Composition

Pipeline(steps)	Pipeline for probabilistic supervised regression.
TransformedTargetRegressor(regressor[, ...])	Transformed target regressor for probabilistic supervised regression.

Model selection and tuning

GridSearchCV(estimator, cv, param_grid[, ...])	Perform grid-search cross-validation to find optimal model parameters.
RandomizedSearchCV(estimator, cv, ...[, ...])	Perform randomized-search cross-validation to find optimal model parameters.

evaluate(estimator, cv, X, y[, scoring, ...])

Evaluate estimator using re-sample folds.

Online learning

OnlineRefit(estimator)	Simple online regression strategy, by refitting the regressor on all data.
OnlineRefitEveryN(estimator[, N])	Online regression strategy, updates only after N new data points are seen.
OnlineDontRefit(estimator)	Simple online regression strategy, turns off any refitting.

OndilOnlineGamlss([distribution, ...])

Wrapper for ondil.online_gamlss.OnlineGamlss.

Reduction - adding predict_proba

This section lists reduction algorithms that take one or multiple sklearn regressors and add a probabilistic prediction mode.

Formally, these algorithms are reduction algorithms, to tabular regression.

BootstrapRegressor(estimator[, ...])

Bootstrap ensemble of a tabular regressor.

ResidualDouble(estimator[, estimator_resid, ...])

Residual double regressor.

MultipleQuantileRegressor([...])

Multiple quantile regressor.

EnbpiRegressor(estimator[, ...])

EnbPI regressor, aka Jackknife+-after-bootstrap, conformal intervals.

MapieRegressor([estimator, method, cv, ...])

MAPIE probabilistic regressor, conformity score based prediction intervals.

MapieSplitConformalRegressor([estimator, ...])	MAPIE Split Conformal Regressor.
MapieCrossConformalRegressor([estimator, ...])	MAPIE Cross Conformal Regressor.
MapieConformalizedQuantileRegressor([...])	MAPIE Conformalized Quantile Regressor.

MapieJackknifeAfterBootstrapRegressor([...])

MAPIE Jackknife+ after Bootstrap Regressor.

BaggingRegressor(estimator[, n_estimators, ...])	Bagging ensemble of probabilistic regresesors.
NGBoostRegressor([dist, score, estimator, ...])	Natural Gradient Boosting Regressor for probabilistic regressors.

CyclicBoosting([feature_groups, ...])

Cyclic boosting regressor from cyclic-boosting library.

Reduction to probabilistic classification

HistBinnedProbaRegressor(clf[, bins])

A binned probabilistic regressor fitting a histogram distribution.

Distributional boosting

XGBoostLSS([dist, stabilization, ...])

Interface to xgboostlss regerssor from the xgboostlss package.

Naive regressors and baselines

This section lists simple regressors which can be used as baselines.

DeltaPointRegressor(estimator)

Delta distribution prediction baseline regressor.

Linear regression

ARDRegression([max_iter, tol, alpha_1, ...])	ARD regression, direct adapter to sklearn ARDRegression.
BayesianRidge([max_iter, tol, alpha_1, ...])	Bayesian ridge regression, direct adapter to sklearn BayesianRidge.
GLMRegressor([family, link, offset_var, ...])	Fits a generalized linear model with a gaussian link.
GlumRegressor([family, link, alpha, ...])	Fits a generalized linear model using the glum package.
PoissonRegressor([alpha, fit_intercept, ...])	Poisson regression, direct adapter to sklearn PoissonRegressor.

Generalized Additive Models

GAMRegressor([terms, distribution, link, ...])

Generalized Additive Model (GAM) Regressor using pygam.

Gaussian process and kernel regression

GaussianProcess([kernel, alpha, optimizer, ...])

Gaussian process (GP), direct interface to sklearn GaussianProcessRegressor.

Bayesian regressors

The below Bayesian regressors provide APIs for prior and posterior handling.

BayesianConjugateLinearRegressor(coefs_prior_cov)	Bayesian probabilistic estimator for linear regression.
BayesianLinearRegressor([prior_config, ...])	Bayesian Linear Regression class with MCMC sampling.

Adapters to other interfaces

SklearnProbaReg(estimator[, inner_type])

Adapter to sklearn regressors with variance prediction interface.

Base classes

BaseProbaRegressor()

Base class for probabilistic supervised regressors.