Resource	GitHub Actions	Code Coverage
Platforms	Windows, macOS, Linux	codecov
R CMD check

CRE

Interpretable Subgroups Identification through Ensemble Learning of Causal Rules

The CRE package provides an interpretable identification of subgroups with heterogeneous causal effects. The heterogeneous subgroups are discovered through ensemble learning of causal rules. Causal rules are highly interpretable if-then statements that recursively partition the feature space into heterogeneous subgroups. A few significant causal rules are selected through Stability Selection to control for family-wise error rate in the finite sample setting. It proposes various estimation methods for the conditional causal effects for each discovered causal rule. It is highly flexible, and multiple causal estimands and imputation methods are implemented.

Installation

Installing from CRAN.

install.packages("CRE")

Installing the latest developing version.

library(devtools)
install_github("NSAPH-Software/CRE", ref="develop")

Import.

library("CRE")

Arguments

Data (required)
y The observed response/outcome vector (binary or continuous).

z The treatment/exposure/policy vector (binary).

X The covariate matrix (binary or continuous).

Parameters (not required)
method_parameters The list of parameters to define the models used, including:

• ratio_dis The ratio of data delegated to the discovery sub-sample (default: 0.5).
• ite_method_dis The method to estimate the individual treatment effect (ITE) on the discovery sub-sample (default: 'aipw') [1].
• ps_method_dis The estimation model for the propensity score on the discovery sub-sample (default: 'SL.xgboost').
• or_method_dis The estimation model for the outcome regressions estimate_ite_aipw on the discovery sub-sample (default: 'SL.xgboost').
• ite_method_inf The method to estimate the individual treatment effect (ITE) on the inference sub-sample (default: 'aipw') [1].
• ps_method_inf The estimation model for the propensity score on the inference subsample (default: 'SL.xgboost').
• or_method_inf The estimation model for the outcome regressions in estimate_ite_aipw on the inference subsample (default: 'SL.xgboost').

hyper_params The list of hyper parameters to finetune the method, including:

• intervention_vars Intervention-able variables used for Rules Generation (default: NULL).
• offset Name of the covariate to use as offset (i.e. 'x1') for T-Poisson ITE Estimation. NULL if not used (default: NULL).
• ntrees_rf A number of decision trees for random forest (default: 20).
• ntrees_gbm A number of decision trees for the generalized boosted regression modeling algorithm. (default: 20).
• node_size Minimum size of the trees' terminal nodes (default: 20).
• max_nodes Maximum number of terminal nodes per tree (default: 5).
• max_depth Maximum rules length (default: 3).
• replace Boolean variable for replacement in bootstrapping (default: TRUE).
• t_decay The decay threshold for rules pruning (default: 0.025).
• t_ext The threshold to define too generic or too specific (extreme) rules (default: 0.01).
• t_corr The threshold to define correlated rules (default: 1).
• t_pvalue The threshold to define statistically significant rules (default: 0.05).
• stability_selection Whether or not using stability selection for selecting the causal rules (default: TRUE).
• cutoff Threshold defining the minimum cutoff value for the stability scores (default: 0.9).
• pfer Upper bound for the per-family error rate (tolerated amount of falsely selected rules) (default: 1).
• penalty_rl Order of penalty for rules length during LASSO for Causal Rules Discovery (i.e. 0: no penalty, 1: rules_length, 2: rules_length^2) (default: 1).

Additional Estimates (not required)
ite The estimated ITE vector. If given, both the ITE estimation steps in Discovery and Inference are skipped (default: NULL).

Notes

[1] Options for the ITE estimation are as follows:

• S-Learner (slearner)
• T-Learner (tlearner)
• T-Poisson (tpoisson)
• X-Learner (xlearner)
• Augmented Inverse Probability Weighting (aipw)
• Causal Forests (cf)
• Bayesian Causal Forests (bcf)
• Bayesian Additive Regression Trees (bart)

if other estimates of the ITE are provided in ite additional argument, both the ITE estimations in discovery and inference are skipped and those values estimates are used instead.

Examples

Example 1 (default parameters)

set.seed(9687)
dataset <- generate_cre_dataset(n = 1000, 
                                rho = 0, 
                                n_rules = 2, 
                                p = 10,
                                effect_size = 2, 
                                binary_covariates = TRUE,
                                binary_outcome = FALSE,
                                confounding = "no")
y <- dataset[["y"]]
z <- dataset[["z"]]
X <- dataset[["X"]]

cre_results <- cre(y, z, X)
summary(cre_results)
plot(cre_results)

Example 2 (personalized ite estimation)

set.seed(9687)
dataset <- generate_cre_dataset(n = 1000, 
                                rho = 0, 
                                n_rules = 2, 
                                p = 10,
                                effect_size = 2, 
                                binary_covariates = TRUE,
                                binary_outcome = FALSE,
                                confounding = "no")
  y <- dataset[["y"]]
  z <- dataset[["z"]]
  X <- dataset[["X"]]

ite_pred <- ... # personalized ite estimation
cre_results <- cre(y, z, X, ite = ite_pred)
summary(cre_results)
plot(cre_results)

Example 3 (setting parameters)

  set.seed(9687)
  dataset <- generate_cre_dataset(n = 1000, 
                                  rho = 0, 
                                  n_rules = 2, 
                                  p = 10,
                                  effect_size = 2, 
                                  binary_covariates = TRUE,
                                  binary_outcome = FALSE,
                                  confounding = "no")
  y <- dataset[["y"]]
  z <- dataset[["z"]]
  X <- dataset[["X"]]

  method_params = list(ratio_dis = 0.25,
                       ite_method_dis="aipw",
                       ps_method_dis = "SL.xgboost",
                       oreg_method_dis = "SL.xgboost",
                       ite_method_inf = "aipw",
                       ps_method_inf = "SL.xgboost",
                       oreg_method_inf = "SL.xgboost")

 hyper_params = list(intervention_vars = c("x1","x2","x3","x4"),
                     offset = NULL,
                     ntrees_rf = 20,
                     ntrees_gbm = 20,
                     node_size = 20,
                     max_nodes = 5,
                     max_depth = 3,
                     t_decay = 0.025
                     t_ext = 0.025,
                     t_corr = 1,
                     t_pvalue = 0.05,
                     replace = FALSE,
                     stability_selection = TRUE,
                     cutoff = 0.8,
                     pfer = 0.1,
                     penalty_rl = 1)

cre_results <- cre(y, z, X, method_params, hyper_params)
summary(cre_results)
plot(cre_results)

More synthetic data sets can be generated using generate_cre_dataset().

Simulations

Discovery.

`CRE/functional_tests/experiments/discovery.R`

Estimation.

`CRE/functional_tests/experiments/estimation.R`

References

Lee, K., Bargagli-Stoffi, F. J., & Dominici, F. (2020). Causal rule ensemble: Interpretable inference of heterogeneous treatment effects. arXiv preprint arXiv:2009.09036.