fit_marg_rule_backfitting: Iteratively back-fit a Super Learner on marginal mixture...

In blind-contours/CVtreeMLE: Cross Validated Decision Trees with Targeted Maximum Likelihood Estimation

Iteratively back-fit a Super Learner on marginal mixture components and covariates

Description

Iteratively back-fit a Super Learner on marginal mixture components and covariates

Usage

fit_marg_rule_backfitting(
  mix_comps,
  at,
  w,
  y,
  w_stack,
  tree_stack,
  fold,
  max_iter,
  verbose,
  parallel_cv,
  seed
)

Arguments

mix_comps	A vector of characters indicating variables for the mixture components
at	Training data
w	A vector of characters indicating variables that are covariates
y	The outcome variable name
w_stack	Stack of algorithms made in SL 3 used in ensemble machine learning to fit Y|W
tree_stack	Stack of algorithms made in SL for the decision tree estimation
fold	Current fold in the cross-validation
max_iter	Max number of iterations of iterative backfitting algorithm
verbose	Run in verbose setting
parallel_cv	Parallelize the cross-validation (TRUE/FALSE)
seed	Numeric, seed number for consistent results

Details

Fit the semi-parametric additive model E(Y) = f(A) + h(W) where f(A) is a Super Learner of decision trees applied to each mixture component and h(W) is a Super Learner applied to the covariates. Each estimator is fit offset by the predictions of the other until convergence where convergence is essentially no difference between the model fits. If a partitioning set is found in f(A) return the rules which are the data-adaptively identified thresholds for the mixture component that maximize the between group difference while controlling for covariates.

Value

A list of the marginal rule results within a fold including:

• marginal_df: A data frame with the data adaptively determined rules found in the partykit model along with the coefficients and other measures.

• models: The best fitting partykit model found for each mixture component in the fold.

Examples

data <- simulate_mixture_cube()
mix_comps <- c("M1", "M2", "M3")
w <- c("age", "sex", "bmi")
sls <- create_sls()
w_stack <- sls$W_stack
tree_stack <- sls$A_stack
example_output <- fit_marg_rule_backfitting(
  mix_comps = mix_comps,
  at = data,
  w = w,
  y = "y",
  w_stack = w_stack,
  tree_stack = tree_stack,
  fold = 1,
  verbose = FALSE,
  parallel_cv = FALSE,
  seed = 6442
)

blind-contours/CVtreeMLE documentation built on June 22, 2024, 8:53 p.m.