guide.GuideBoostingRegressor

class guide.GuideBoostingRegressor(*, loss='squared_error', n_estimators=100, learning_rate=0.1, max_depth=3, min_samples_leaf=2, min_child_weight=1.0, reg_lambda=1.0, gamma=0.0, subsample=1.0, max_features=None, early_stopping_rounds=None, interactions=True, significance_level=0.1, max_split_candidates=100, categorical_features=None, feature_roles=None, periodic_features=None, missing_flags=None, n_jobs=None, random_state=0)

Gradient-boosted GUIDE regression trees (experimental).

Boosting is a guide-rs extension, not part of GUIDE; its variable selection and defaults may change between releases.

Each split variable is chosen by chi-square tests of the gradients against each predictor’s bins: GUIDE’s tests on their signs for squared error, and score tests for bin effects for the Poisson and logistic losses. Boosted trees thus inherit GUIDE’s unbiased variable selection; the split point maximizes the second-order (Newton) gain.

Parameters:

loss{“squared_error”, “poisson”, “logistic”}, default=”squared_error”

Poisson uses a log link and accepts exposure; logistic fits a 0/1 response and predicts probabilities.

n_estimatorsint, default=100

Maximum number of boosting rounds. Multiclass models fit one tree per class in each round.

learning_ratefloat, default=0.1

Shrinkage applied to every leaf value.

max_depthint, default=3

Depth of each tree; boosted trees are not pruned.

min_samples_leafint, default=2

min_child_weightfloat, default=1.0

Minimum sum of loss Hessians (times sample weights) in a child.

reg_lambdafloat, default=1.0

L2 penalty on leaf values.

gammafloat, default=0.0

Minimum gain to split.

subsamplefloat, default=1.0

Fraction of rows sampled without replacement for each round.

max_featuresint or float, default=None

Split variables sampled at each node; by default all.

early_stopping_roundsint, default=None

Stop when the loss on validation_data (passed to fit) has not improved for this many rounds, and keep the best round.

interactions, significance_level, max_split_candidates, categorical_features, feature_roles, periodic_features, missing_flags, n_jobs, random_state

As in GuideRegressor. After the first round the chi-square statistics rank variables; they are not significance tests.

__init__(*, loss='squared_error', n_estimators=100, learning_rate=0.1, max_depth=3, min_samples_leaf=2, min_child_weight=1.0, reg_lambda=1.0, gamma=0.0, subsample=1.0, max_features=None, early_stopping_rounds=None, interactions=True, significance_level=0.1, max_split_candidates=100, categorical_features=None, feature_roles=None, periodic_features=None, missing_flags=None, n_jobs=None, random_state=0)

Methods

__init__(*[, loss, n_estimators, ...])
apply(X)	Terminal node ids, shape (n_samples, n_trees).
fit(X, y[, sample_weight, exposure, ...])	Fit the model.
get_metadata_routing()	Get metadata routing of this object.
get_params([deep])	Get parameters for this estimator.
predict(X[, exposure])	Predicted response; Poisson models fitted with exposure need it here too.
score(X, y[, sample_weight])	Return coefficient of determination on test data.
set_fit_request(*[, exposure, ...])	Configure whether metadata should be requested to be passed to the fit method.
set_params(**params)	Set the parameters of this estimator.
set_predict_request(*[, exposure])	Configure whether metadata should be requested to be passed to the predict method.
set_score_request(*[, sample_weight])	Configure whether metadata should be requested to be passed to the score method.
to_dot()	The fitted model in Graphviz DOT format.
to_json()	The fitted model in the format saved by the guide-rs CLI.
to_text()	The fitted model as indented text.

Attributes

n_estimators_

Boosting rounds kept; with early stopping, the best validation round.