VineRegressor.__init__

VineRegressor.__init__(mean=True, quantiles=None, backend=None, batch_size=100, use_grid=True, normalize_weights=True, random_state=None)

Sklearn-compatible vine-copula regressor.

Predicts the conditional mean \(\hat{\mathbb{E}}[Y \mid X = x]\) and/or conditional quantiles using the weighted-sample estimator derived in the class docstring.

Parameters:

meanbool, default=True

If True, predict the conditional mean. Set to False to get quantile-only predictions (quantiles must then be set).

quantilesarray-like of float, shape (n_quantiles,), default=None

Quantile levels in (0, 1) to predict. None disables quantile prediction.

backendVinecopBackend or compatible, default=None

Backend instance bundling fit-time controls and an optional pre-specified structure on (Y, X_1, ..., X_d) (Y always in the first dimension). None resolves to a default VinecopBackend with the tll pair family at fit time.

batch_sizeint, default=100

Number of test points processed per batch in predict.

use_gridbool, default=True

Controls how training responses are represented for the weighted-sample predictor. False uses importance weighting over training rows with \(w_i(x) \propto c_{Y,X}(\hat F_Y(y_i), \hat F_X(x))\). True (default) uses the Kde1d grid points and an extra \(\hat f_Y(y_g)\) factor.

normalize_weightsbool, default=True

If True (default), per-row weights produced by _iter_weights are normalised to sum to one. Forest wrappers set this to False so they can average raw weights across trees and normalise once at the ensemble level.

random_stateint, RandomState instance or None, default=None

Seeds the RNG used by stochastic operations. Resolved via sklearn.utils.check_random_state inside fit.