Create an Ensemble of CATE EP-learners with Varying Sieve Basis Sizes

This function creates an ensemble of CATE EP-learners of varying sieve dimensions for use with cross-validation.

Usage

make_ep_stack(
  base_learner,
  hte3_task,
  treatment_level = 1,
  control_level = 0,
  sieve_basis_grid = NULL,
  targeting_style = "dr",
  r_targeting_basis = "v_plus_propensity"
)

Arguments

base_learner

The base learner of Lrnr_cate_EP.

hte3_task

A hte3_Task object containing the data and necessary information for the heterogeneous treatment effect estimation.

treatment_level

Treatment level used for the treated arm in each EP learner in the stack.

control_level

Reference treatment level used for the control arm in each EP learner in the stack.

sieve_basis_grid

Optional vector of sieve basis sizes to include in the ensemble. If NULL, the default heuristic grid uses the first-stage EP sieve dimension for the chosen targeting style.

targeting_style

EP targeting style or styles to include in the stack. Use c("dr", "r") to build both EP variants across the requested basis grid.

r_targeting_basis

First-stage basis construction used for EP-R fits in the stack.

Value

A stack of CATE EP meta-learners with varying sieve basis sizes.

Examples

if (FALSE) { # \dontrun{
library(sl3)

data <- hte3_example_data(n = 80, seed = 1)
task <- hte_task(
  data = data,
  modifiers = c("W1", "W2"),
  confounders = c("W1", "W2", "W3"),
  treatment = "A",
  outcome = "Y",
  propensity_learner = Lrnr_mean$new(),
  outcome_learner = Lrnr_mean$new(),
  mean_learner = Lrnr_mean$new(),
  cross_fit = FALSE
)

ep_stack <- make_ep_stack(
  base_learner = Lrnr_mean$new(),
  hte3_task = task,
  sieve_basis_grid = c(2, 4, 6),
  targeting_style = c("dr", "r")
)

ep_stack
} # }