Skip to contents

The Super Learner ———————————————

Source: R/flxipt_native.R
super_learner.Rd

A step of the flxipt_native implementation. Called by flexible_iptw. See the file header for the source the module follows.

Usage

super_learner(
  y,
  X,
  library = NULL,
  n_folds = 10,
  meta = "nnls",
  binary = NULL,
  loss = "l2",
  ridge = 1e-08,
  honest_level_one = TRUE
)

Arguments

y

Coerced to numeric by the body, with as.numeric.

X

Optional; may be NULL. A matrix; passed to as.matrix.

library

The body requires: flxipt: the library is empty.

n_folds

Passed to .flxipt_folds. Defaults to 10.

meta

One of "discrete", "nnls". Defaults to "nnls".

binary

Optional; may be NULL. A flag; the body branches on it.

loss

Compared against "nll". Defaults to "l2".

ridge

Numeric; passed to max. Defaults to 1e-08.

honest_level_one

A flag; the body branches on it. Defaults to TRUE.

Value

A list with fitted, estimate, weights, weight_vector, cv_risk, cv_risk_ensemble, best_candidate, best_candidate_risk, discrete_choice, level_one, candidate_fits, library, n, n_folds, meta, loss, binary, honest_level_one, method.

Examples

set.seed(1)
n <- 40
X <- matrix(rnorm(n * 2), n, 2)
y <- 1 + X[, 1] - 0.5 * X[, 2] + rnorm(n, 0, 0.3)
r <- super_learner(y, X, n_folds = 4)
str(r, max.level = 1)
#> List of 19
#>  $ fitted             : num [1:40] 0.5346 1.3092 -0.0221 2.2486 1.6271 ...
#>  $ estimate           : num 0.0602
#>  $ weights            : Named num [1:6] 0 1 0 0 0 0
#>   ..- attr(*, "names")= chr [1:6] "intercept" "main" "quadratic" "interaction" ...
#>  $ weight_vector      : num [1:6] 0 1 0 0 0 0
#>  $ cv_risk            : Named num [1:6] 0.7277 0.0602 0.0668 0.0644 0.068 ...
#>   ..- attr(*, "names")= chr [1:6] "intercept" "main" "quadratic" "interaction" ...
#>  $ cv_risk_ensemble   : num 0.0602
#>  $ best_candidate     : chr "main"
#>  $ best_candidate_risk: num 0.0602
#>  $ discrete_choice    : chr "main"
#>  $ level_one          : num [1:40, 1:6] 1.078 1.083 0.964 1.143 1.078 ...
#>  $ candidate_fits     : num [1:40, 1:6] 1.07 1.07 1.07 1.07 1.07 ...
#>  $ library            : chr [1:6] "intercept" "main" "quadratic" "interaction" ...
#>  $ n                  : int 40
#>  $ n_folds            : int 4
#>  $ meta               : chr "nnls"
#>  $ loss               : chr "l2"
#>  $ binary             : logi FALSE
#>  $ honest_level_one   : logi TRUE
#>  $ method             : chr "Super Learner, van der Laan, Polley & Hubbard (2007) Sec. 2 eq. (1)"