Train a PerpetualBooster model — perpetual • perpetual

Perpetual is a self-generalizing gradient boosting machine that doesn't need hyperparameter optimization. It automatically finds the best configuration based on the provided budget.

Usage

perpetual(
  x,
  y,
  objective = "LogLoss",
  budget = NULL,
  iteration_limit = NULL,
  stopping_rounds = NULL,
  max_bin = NULL,
  num_threads = NULL,
  missing = NULL,
  allow_missing_splits = NULL,
  create_missing_branch = NULL,
  missing_node_treatment = NULL,
  log_iterations = NULL,
  quantile = NULL,
  reset = NULL,
  timeout = NULL,
  memory_limit = NULL,
  seed = NULL,
  ...
)

Arguments

x: A matrix or data.frame of features.
y: A vector of targets (numeric for regression, factor/integer for classification).
objective: A string specifying the objective function. Default is "LogLoss".
budget: A numeric value ensuring the training time does not exceed this budget (in normalized units).
iteration_limit: An integer limit on the number of iterations.
stopping_rounds: An integer for early stopping.
max_bin: Integer, max number of bins for histograms.
num_threads: Integer, number of threads to use.
missing: Value to consider as missing data. Default is NaN.
allow_missing_splits: Boolean.
create_missing_branch: Boolean. Whether to create a separate branch for missing values (ternary trees).
missing_node_treatment: String. How to handle weights for missing nodes if create_missing_branch is True. Options: "None", "AssignToParent", "AverageLeafWeight", "AverageNodeWeight".
log_iterations: Integer.
quantile: Numeric.
reset: Boolean.
timeout: Numeric.
memory_limit: Numeric.
seed: Integer seed for reproducibility.
...: Additional arguments.

Value

A PerpetualBooster object.