Linear regression objective — objective

Builds a squared-error regression objective with optional elastic net penalization.

Usage

objective_linear(
  x,
  y,
  weights = NULL,
  lambda = 0,
  alpha = 0,
  lambda_l1 = NULL,
  lambda_l2 = NULL,
  include_intercept = TRUE,
  penalize_intercept = FALSE,
  loss_scale = NULL,
  id = NULL,
  label = NULL,
  transform = objective_transform_identity()
)

Arguments

x: (matrix()) Design matrix without an intercept column. Rows correspond to observations.
y: (numeric()) Numeric responses for the regression task.
weights: (numeric()) Optional non-negative weights with length nrow(x).
lambda: (numeric(1)) Overall penalty strength used together with alpha for elastic net penalties.
alpha: (numeric(1)) Elastic net mixing parameter in [0, 1]. 0 corresponds to ridge, 1 to lasso.
lambda_l1: (numeric(1)) Optional direct specification of the L1 penalty weight. Overrides lambda * alpha when provided.
lambda_l2: (numeric(1)) Optional direct specification of the L2 penalty weight. Overrides lambda * (1 - alpha) when provided.
include_intercept: (logical(1)) If TRUE, an intercept column is prepended to x before constructing the objective.
penalize_intercept: (logical(1)) If FALSE and include_intercept = TRUE, the intercept term is excluded from the penalties.
loss_scale: (numeric(1)) Scaling factor applied to the summed squared errors. Defaults to 1 / (2 * sum(weights)).
id: (character(1)) Identifier forwarded to Objective. Defaults to "logreg".
label: (character(1)) Label used for the resulting objective. Defaults to "logistic risk".
transform: (objective_transform()) Optional scalar transformation applied to the empirical risk before it is returned. Defaults to objective_transform_identity().

Value

An Objective instance capturing the squared-error regression risk.