Skip to contents

Recompute Policy Trees with Custom Parameters

Source: R/margot_policy_tree.R
margot_policy_tree.Rd

Computes policy trees for causal forest models with flexible covariate selection and train/test split options. This function provides a direct way to generate policy trees without running full causal forest analysis, paralleling the functionality of margot_rate() and margot_qini().

Usage

margot_policy_tree(
  model_results,
  model_names = NULL,
  custom_covariates = NULL,
  exclude_covariates = NULL,
  covariate_mode = c("original", "custom", "add", "all"),
  depth = "both",
  train_proportion = 0.5,
  label_mapping = NULL,
  verbose = TRUE,
  seed = 12345,
  tree_method = c("policytree", "fastpolicytree")
)

Arguments

model_results

List returned by margot_causal_forest() or margot_flip_forests(), containing results and optionally covariates and data.

model_names

Optional character vector specifying which models to process. Default NULL (all models).

custom_covariates

Character vector of covariate names to use for policy trees. If NULL, uses the original top variables from the model.

exclude_covariates

Character vector of covariate names or patterns to exclude. Supports exact matches and regex patterns (e.g., "_log" excludes all variables containing "_log").

covariate_mode

Character string specifying how to handle covariates:

  • "original"Use original top variables from model (default)

  • "custom"Use only the specified custom_covariates

  • "add"Add custom_covariates to existing top variables

  • "all"Use all available covariates

depth

Numeric or character specifying which depth(s) to compute: 1 for single split, 2 for two splits, or "both" for both depths (default).

train_proportion

Numeric value between 0 and 1 for the proportion of data used for training depth-2 trees. Default is 0.5. Note: depth-1 trees use all available data but only the selected covariates (same as depth-2).

label_mapping

Named character vector for converting variable names to readable labels.

verbose

Logical; print progress messages (default TRUE).

seed

Integer; base seed for reproducible computations (default 12345).

tree_method

Character string specifying the package to use: "policytree" (default) or "fastpolicytree". The fastpolicytree package provides ~10x faster computation with identical results. Falls back to policytree if fastpolicytree is not installed.

Value

A list structured similarly to margot_causal_forest() output, containing:

  • results: List where each element corresponds to a model and contains:

    • dr_scores: Doubly robust scores (original or flipped if available)

    • policy_tree_depth_1: Single-split policy tree (if requested)

    • policy_tree_depth_2: Two-split policy tree (if requested and possible)

    • plot_data: Data for visualization (X_test, X_test_full, predictions)

    • top_vars: Variables used for policy trees

    • policy_tree_covariates: Final covariate selection

    • policy_tree_metadata: Metadata about the computation

  • covariates: The covariate matrix used

  • not_missing: Indices of complete cases

  • train_proportion: The train/test split proportion used

Details

This function allows you to:

  • Exclude specific covariates (e.g., log-transformed variables with "_log")

  • Use custom covariate sets for policy optimization

  • Adjust the train/test split for depth-2 trees

  • Recompute policy trees without re-running causal forests

The output is structured to be compatible with margot_policy(), margot_plot_policy_tree(), margot_plot_policy_combo(), and margot_interpret_policy_tree().

Examples

if (FALSE) { # \dontrun{
# Recompute policy trees with default settings
policy_trees <- margot_policy_tree(causal_forest_results)

# Exclude log-transformed variables
policy_trees_no_log <- margot_policy_tree(
  causal_forest_results,
  exclude_covariates = "_log"
)

# Use custom covariates with 80/20 train/test split
policy_trees_custom <- margot_policy_tree(
  causal_forest_results,
  custom_covariates = c("age", "gender", "income"),
  covariate_mode = "custom",
  train_proportion = 0.8
)

# Compute only depth-1 trees using all covariates
policy_trees_d1 <- margot_policy_tree(
  causal_forest_results,
  covariate_mode = "all",
  depth = 1
)

# Process specific models with larger training set
policy_trees_selected <- margot_policy_tree(
  causal_forest_results,
  model_names = c("anxiety", "depression"),
  train_proportion = 0.8
)

# Visualize results
plot <- margot_plot_policy_tree(policy_trees, "model_anxiety")

# Create combined plots
plots <- margot_plot_policy_combo(policy_trees, "model_anxiety")
print(plots$combined_plot)

# Interpret results
margot_interpret_policy_tree(policy_trees, "model_anxiety")
} # }