Collection of hyperparameters specific to LightGBM. Each can be
passed to set_engine for tuning. See the LightGBM
docs for
more information on each parameter.
max_bin(range = c(100L, 500L), trans = NULL)
max_depth(range = c(3L, 17L), trans = NULL)
add_to_linked_depth(range = c(1L, 3L), trans = NULL)
learning_rate(range = c(-3, -0.5), trans = scales::log10_trans())
feature_fraction(range = c(0.3, 1), trans = NULL)
feature_fraction_bynode(range = c(0.3, 1), trans = NULL)
min_data_in_leaf(range = c(20L, 500L), trans = NULL)
min_gain_to_split(range = c(-3, 2), trans = scales::log10_trans())
num_leaves(range = c(31L, 4000L), trans = NULL)
bagging_fraction(range = c(0.3, 1), trans = NULL)
bagging_freq(range = c(1L, 100L), trans = NULL)
min_data_per_group(range = c(20L, 250L), trans = NULL)
max_cat_threshold(range = c(20L, 250L), trans = NULL)
lambda_l1(range = c(-3, 2), trans = scales::log10_trans())
lambda_l2(range = c(-3, 2), trans = scales::log10_trans())
cat_smooth(range = c(10, 100), trans = NULL)
cat_l2(range = c(-3, 2), trans = scales::log10_trans())A two-element vector holding the defaults for the smallest and largest possible values, respectively.
A trans object from the scales package, such
as log10_trans or
reciprocal_trans. If not provided, the default is
used which matches the units used in range. If no transformation, NULL.
max_depth()
#> Max Tree Depth (quantitative)
#> Range: [3, 17]
learning_rate()
#> Learning Rate (quantitative)
#> Transformer: log-10 [1e-100, Inf]
#> Range (transformed scale): [-3, -0.5]
feature_fraction()
#> % Features Selected by Iteration (quantitative)
#> Range: [0.3, 1]
lambda_l1()
#> L1 Regularization (quantitative)
#> Transformer: log-10 [1e-100, Inf]
#> Range (transformed scale): [-3, 2]
lambda_l2()
#> L2 Regularization (quantitative)
#> Transformer: log-10 [1e-100, Inf]
#> Range (transformed scale): [-3, 2]
cat_smooth()
#> Categorical Smoothing (quantitative)
#> Range: [10, 100]