boosters.GBLinearConfig#

class boosters.GBLinearConfig#

Bases: object

Main configuration for GBLinear model.

GBLinear uses gradient boosting to train a linear model via coordinate descent. Simpler than GBDT but can be effective for linear relationships.

Parameters:

n_estimators – Number of boosting rounds. Default: 100.
learning_rate – Step size for weight updates. Default: 0.5.
objective – Loss function for training. Default: Objective.Squared().
metric – Evaluation metric. None uses objective’s default.
l1 – L1 regularization (alpha). Encourages sparse weights. Default: 0.0.
l2 – L2 regularization (lambda). Prevents large weights. Default: 0.0.
early_stopping_rounds – Stop if no improvement for this many rounds.
seed – Random seed for reproducibility. Default: 42.

Examples

>>> config = GBLinearConfig(
...     n_estimators=200,
...     learning_rate=0.3,
...     objective=Objective.logistic(),
...     l2=0.1,
... )

classmethod __new__(*args, **kwargs)#

early_stopping_rounds#: Early stopping rounds (None = disabled).

l1#: L1 regularization (alpha).

l2#: L2 regularization (lambda).

learning_rate#: Learning rate (step size).

max_delta_step#

Maximum per-coordinate Newton step (stability), in absolute value.

Set to 0.0 to disable.

metric#: Get the evaluation metric (or None).

n_estimators#: Number of boosting rounds.

objective#: Get the objective function.

seed#: Random seed.

update_strategy#: Coordinate descent update strategy.

verbosity#: Verbosity level for training output.