LightGBMHyperparameters

Properties

max_nrounds

integerdefault 1000≥ 1

Maximum number of boosting rounds when tuning nrounds by early stopping.

force_nrounds

integer | nulldefault null≥ 1

Use this many boosting rounds; disables the search for nrounds.

early_stopping_rounds

integerdefault 10≥ 1

Number of rounds without improvement to stop training.

num_leavestunable

integerdefault 8≥ 1

Maximum number of leaves in one tree.

max_depthtunable

integerdefault -1

Maximum tree depth. -1 = no limit.

learning_ratetunable

numberdefault 0.01≤ 1> 0

Learning rate.

feature_fractiontunable

numberdefault 1≤ 1> 0

Fraction of features sampled per tree.

subsampletunable

numberdefault 1≤ 1> 0

Fraction of cases sampled per tree (bagging fraction).

subsample_freqtunable

integerdefault 1≥ 1

Bagging frequency.

lambda_l1tunable

numberdefault 0≥ 0

L1 regularization.

lambda_l2tunable

numberdefault 0≥ 0

L2 regularization.

max_cat_thresholdtunable

integerdefault 32≥ 1

Maximum number of split points for categorical features.

min_data_per_grouptunable

integerdefault 32≥ 1

Minimum number of cases per categorical group.

linear_treetunable

booleandefault false

Fit linear models at leaves.

ifwtunable

booleandefault false

Inverse Frequency Weighting in classification.

objective

string | nulldefault null

LightGBM objective. NULL = set from outcome type.

device_type

stringdefault "cpu"

Compute device.

one of"cpu""gpu""cuda"

tree_learner

stringdefault "serial"

Tree learner type.

one of"serial""feature""data""voting"

force_col_wise

booleandefault true

Force column-wise histogram building (CPU only).

Relationships