Interpret Policy Tree Results
Source:R/margot_interpret_policy_tree.R
margot_interpret_policy_tree.RdThis function creates an interpretation of policy tree results from a causal forest or multi-arm causal forest model. It generates a formatted description of the policy tree, including the main splits and recommended actions.
Usage
margot_interpret_policy_tree(
model,
model_name,
max_depth = 2L,
train_proportion = 0.5,
custom_action_names = NULL,
label_mapping = NULL,
original_df = NULL,
remove_tx_prefix = TRUE,
remove_z_suffix = TRUE,
use_title_case = TRUE,
include_conditional_means = TRUE,
use_math_notation = FALSE,
output_format = c("prose", "bullet"),
report_policy_value = c("none", "treat_all", "control_all", "both", "treated_only"),
policy_value_R = 499L,
policy_value_seed = 42L,
policy_value_ci_level = 0.95
)Arguments
- model
A list containing the results from a multi-arm causal forest model.
- model_name
A string specifying which model's results to interpret.
- max_depth
Integer, 1 or 2; which stored tree to interpret.
- train_proportion
Numeric value between 0 and 1 for the proportion of data used for training. Default is 0.5.
- custom_action_names
Optional vector of custom names for the actions. Must match the number of actions in the policy tree.
- label_mapping
Optional list that maps variable names to custom labels.
- original_df
Optional dataframe with untransformed variables, used to display split values on the data scale.
- remove_tx_prefix
Logical indicating whether to remove prefixes like t0_ from variable names. Default is TRUE.
- remove_z_suffix
Logical indicating whether to remove the _z suffix from variable names. Default is TRUE.
- use_title_case
Logical indicating whether to convert variable names to title case. Default is TRUE.
- include_conditional_means
Logical indicating whether to include conditional means information if available. Default is TRUE.
- use_math_notation
Logical indicating whether to use mathematical notation (E[Y(a)|X]) or plain language. Default is FALSE for clarity.
- output_format
Character string specifying output format: "prose" (default) for flowing narrative text, or "bullet" for structured bullet points.
- report_policy_value
Character: one of "none" (default), "treat_all", "control_all", or "both". If not "none", appends policy value summary with 95 stored in `plot_data` and DR scores from the model.
- policy_value_R
Integer ≥ 199; number of bootstrap replicates (default 499).
- policy_value_seed
Integer or NULL; RNG seed (default 42).
- policy_value_ci_level
Numeric confidence level (default 0.95).