RangerHyperparameters

Properties

num_treestunable

integer≥ 1

Number of trees.

mtry

integer | integer[] | null≥ 1≥ 1 items

Number of features considered at each split. NULL = ranger default. Cannot exceed the number of features in the training data.

importance

string

Variable importance mode.

one of"none""impurity""impurity_corrected""permutation"

write_forest

boolean

Save the forest object (required for prediction).

probability

boolean

Grow a probability forest (classification only).

min_node_size

integer | integer[] | null≥ 1≥ 1 items

Minimal node size. NULL = ranger default by task type.

min_bucket

integer | null≥ 1

Minimal number of samples in a terminal node (survival only).

max_depth

integer | integer[] | null≥ 0≥ 1 items

Maximal tree depth. NULL or 0 = unlimited.

replacetunable

boolean

Sample with replacement.

sample_fractiontunable

number≤ 1> 0

Fraction of observations to sample per tree.

case_weights

array | null≥ 1 items

Per-observation sampling weights. Must have one value per case.

class_weights

array | null≥ 1 items

Per-class weights (classification only). Must have one value per class.

splitrule

string | string[] | null≥ 1 items

Splitting rule (task-dependent). NULL = ranger default.

num_random_splitstunable

integer≥ 1

Random splits per candidate variable (extratrees splitrule).

alphatunable

number≥ 0≤ 1

Significance threshold to allow splitting (maxstat splitrule).

minproptunable

number≥ 0≤ 1

Lower quantile of covariate distribution considered for splitting (maxstat splitrule).

poisson_tau

number> 0

Tau parameter (poisson splitrule).

split_select_weights

number[] | number[][] | null≥ 1 items≥ 1 items

Per-feature split-selection probabilities: one vector applied to every tree, or one vector per tree.

always_split_variables

array | null≥ 1 items

Variables always included as split candidates. Values must name training features.

respect_unordered_factors

string | null

Handling of unordered factors. NULL uses the ranger default.

one of"partition""ignore""order"

scale_permutation_importance

boolean

Scale permutation importance by its standard error.

local_importance

boolean

Compute local (per-observation) permutation importance.

regularization_factortunable

number≥ 0

Regularization factor penalizing variables with many split points.

regularization_usedepth

boolean

Apply the regularization factor with node depth.

keep_inbag

boolean

Record how often each observation is in-bag per tree.

inbag

array | null≥ 1 items

Manually set in-bag counts: one per-case count vector per tree.

holdout

boolean

Hold-out mode: hold out samples with case weight 0.

quantreg

boolean

Prepare quantile prediction (regression only).

time_interest

array | null≥ 1 items

Time points of interest for survival prediction.

oob_error

boolean

Compute the OOB prediction error.

save_memory

boolean

Use the memory-saving (slower) splitting mode.

verbose

boolean

Show ranger computation status.

node_stats

boolean

Save additional node statistics.

seed

integer | null

Random seed. NULL = generated from R.

na_action

string

How to handle missing values.

one of"na.learn""na.omit""na.fail"

ifwtunable

boolean

Inverse Frequency Weighting in classification.

Relationships