control_owl {polle}

Control arguments for Outcome Weighted Learning

Description

control_owl() sets the default control arguments for backwards outcome weighted learning, type = "owl". The arguments are passed directly to DTRlearn2::owl() if not specified otherwise.

Usage

control_owl(
  policy_vars = NULL,
  reuse_scales = TRUE,
  res.lasso = TRUE,
  loss = "hinge",
  kernel = "linear",
  augment = FALSE,
  c = 2^(-2:2),
  sigma = c(0.03, 0.05, 0.07),
  s = 2^(-2:2),
  m = 4
)

Arguments

policy_vars

Character vector/string or list of character vectors/strings. Variable names used to restrict the policy. The names must be a subset of the history names, see get_history_names(). Not passed to owl().

reuse_scales

The history matrix passed to owl() is scaled using scale() as advised. If TRUE, the scales of the history matrix will be saved and reused when applied to (new) test data.

res.lasso

If TRUE a lasso penalty is applied.

loss

Loss function. The options are "hinge", "ramp", "logit", "logit.lasso", "l2", "l2.lasso".

kernel

Type of kernel used by the support vector machine. The options are "linear", "rbf".

augment

If TRUE the outcomes are augmented.

c

Regularization parameter.

sigma

Tuning parameter.

s

Slope parameter.

m

Number of folds for cross-validation of the parameters.

Value

list of (default) control arguments.