Task object for meta-learners in the point-treatment setting.

Constructs a hte3_Task object for meta-learners in the point-treatment setting containing relevant data and nuisance function estimators.

Usage

make_hte3_Task_tx(
  data,
  modifiers,
  confounders,
  treatment,
  outcome,
  id = NULL,
  weights = NULL,
  treatment_type = c("default", "binomial", "categorical", "continuous"),
  pi.hat = NULL,
  mu.hat = NULL,
  m.hat = NULL,
  learner_pi = get_autoML(),
  learner_mu = get_autoML(),
  learner_m = NULL,
  multinomial_learner = Lrnr_independent_binomial,
  cross_fit_and_cv = TRUE,
  folds = 10,
  warn = TRUE,
  for_prediction = FALSE,
  ...
)

Arguments

data: A named data frame or data.table containing treatment effect modifiers, potential confounders, treatment, and outcome. Optionally, the dataset can contain weights and subject IDs. See the data argument of sl3_Task for further options.
modifiers: A character vector of variable names in colnames(data) for treatment effect moderators.
confounders: A character vector of variable names in colnames(data) for potential confounders W for which to adjust.
treatment: A character specifying the variable name in colnames(data) for the numeric treatment assignment A.
outcome: A character specifying the variable name in colnames(data) for the outcome variable Y.
id: An (optional) character specifying the variable name in colnames(data) for observation IDs.
weights: An (optional) character specifying the variable name in colnames(data) for observation weights.
pi.hat: An (optional) numeric matrix of dimension n by nlevels(A) containing estimates of the propensity score a -> pi(a | W_i) at each treatment level, where column j corresponds to treatment level sort(unique(A))[j]. This argument can be used by DR-type learners. Alternatively, the estimates can be learned internally by passing an sl3_Learner to the learner_pi argument.
mu.hat: An (optional) numeric matrix of dimension n by nlevels(A) containing estimates of the outcome regression a -> mu(a, W_i) at each treatment level, where column j corresponds to treatment level sort(unique(A))[j]. This argument can be used by DR-type learners. Alternatively, the estimates can be learned internally by passing an sl3_Learner to the learner_m argument.
m.hat: An (optional) numeric vector of size n containing estimates of the conditional mean outcome E[Y | W_i]. This argument can be used by R-type learners. Alternatively, the estimates can be learned internally by passing an sl3_Learner to the learner_m argument.
learner_pi: A binomial sl3_Learner or Stack object specifying the learning algorithm to estimate the propensity score w -> pi(a_star | w) at a given treatment level a_star. During training, this learner is passed an sl3_Task object task that contains a feature matrix task$X of confounders and an outcome vector task$Y corresponding to the binary treatment indicator 1(A = a_star). By default, cross_fit_and_cv = TRUE, and learner_m is fit with 10-fold cross-fitting using make_cross_fitted. If a Stack object, the best model is selected using cross-validation.
learner_mu: A sl3_Learner or sl3_Task object specifying the learning algorithm to estimate the outcome regression w -> mu(a_star, w) at a given treatment level a_star. The outcome regression is estimated via a treatment-stratified regression. During training, this learner is passed an sl3_Task object task that is subsetted to variables with treatment A = a_star, and contains a feature matrix task$X of confounders and an outcome vector task$Y corresponding to outcome. By default, cross_fit_and_cv = TRUE, and learner_m is fit with 10-fold cross-fitting using make_cross_fitted. If a Stack object, the best model is selected using cross-validation.
learner_m: A sl3_Learner or sl3_Task object specifying the learning algorithm to estimate the conditional mean outcome m. During training, this learner is passed an sl3_Task object task that contains a feature matrix task$X of confounders and an outcome vector task$Y corresponding to outcome. By default, cross_fit_and_cv = TRUE, and learner_m is fit with 10-fold cross-fitting using make_cross_fitted. If a Stack object, the best model is selected using cross-validation.
multinomial_learner: A multinomial Lrnr_base object to convert learner_pi from a binomial learner to a multinomial learner.
cross_fit_and_cv: Whether to cross-fit the specified nuisance learners by applying learner <- make_cross_fitted(learner).
for_prediction: A boolean of whether to return an hte3_Task without the nuisance training and estimates. This can be useful when you wish to construct a task for predicting on a new dataset.
...: Additional arguments to pass to the sl3_Task and tmle_Task constructors.

Value

A hte3_Task object for the point-treatment data-structure