Finding Subgroups in Treatment Effects with Conformal Trees

Usage

r2p_hte(
  data,
  target,
  treatment,
  learner,
  cv_folds = 10,
  alpha = 0.05,
  gamma = 0.1,
  lambda = 0.5,
  max_groups = 10
)

Arguments

data: (data.frame)
data set for model training and uncertainty estimation.
target: (string)
name of the target variable. The target must be a numeric variable.
treatment: (string)
name of the treatment variable. If treatment
is a factor, then the first level is treated as control and the second level as treatment indicator. If treatment
is a numeric, then zero-one encoding is assumed and "1"
treated as treatment indicator.
learner: (model_spec)
the learner for training the prediction model. See parsnip::model_spec() for details.
cv_folds: (count)
number of CV+ folds.
alpha: (proportion)
miscoverage rate.
gamma: (proportion)
regularization parameter ensuring that reduction in the impurity of the confident homogeneity is sufficiently large.
lambda: (proportion)
balance parameter, quantifying the impact of the average interval length relative to the average absolute deviation (i.e. interval size vs. within-group homogeneity)
max_groups: (count)
maximum number of subgroups.

Value

The tree.

Examples

library(tidymodels)
library(ranger)
data(bikes)
set.seed(1234)
randforest <- rand_forest() %>%
  set_mode("regression") %>%
  set_engine("ranger")
groups <- r2p_hte(
  data = bikes,
  target = "count",
  treatment = "year",
  learner = randforest,
  cv_folds = 10,
  alpha = 0.05,
  gamma = 0.2,
  lambda = 0.5,
  max_groups = 10
)
#> Error in n + 1: non-numeric argument to binary operator
groups$tree
#> Error in groups$tree: object of type 'closure' is not subsettable