Grow Linear (Dart) Trees — grow_linear

LightGBM Piecewise-Linear DART: fits a ridge regression at each leaf (rather than a constant) while applying tree dropout to prevent early linear models from dominating later rounds. Inspired by GBDT-PL https://arxiv.org/abs/1802.05640 with DART regularization.

Usage

grow_linear_trees(
  x,
  y,
  trees = 100L,
  task = "classification",
  drop_rate = 0.1,
  linear_lambda = 0.5,
  exotic = FALSE,
  ...
)

Arguments

x: a data X
y: a data Y
trees: num of boosting rounds
task: a task
drop_rate: fraction of previous trees dropped per round
linear_lambda: L2 regularization on leaf linear models
exotic: logical; if TRUE, apply additional exotic parameters
...: other eng args

Value

a fitted model

Details

When exotic = FALSE (default), parameters follow the GBDT-PL paper spirit: linear leaves, DART dropout, forest normalization, path smoothing for small-leaf stability.

When exotic = TRUE, ExtraTrees-style random split thresholds are added (extra_trees = TRUE), creating triple stochasticity: random thresholds + column subsampling + tree dropout. one_drop = TRUE guarantees at least one tree is always dropped.