Bayesian nonparametric mixtures of categorical directed graphs for heterogeneous causal inference

This repository contains the R codes implementing DP mixture of categorical DAGs from arxiv.

MCMC

The folder MCMC contains the R codes for the implementation of the main MCMC algorithms for the DP mixture of categorical DAGs. \ In particular:

GIBBS_collapsed_rcpp.R : contains the main MCMC algorithm for posterior inference
move_dag.R : implements the proposal distribution of DAGs
sample_from_baseline.R : samples from the baseline over the space of DAGs
marg_dag.R : computes the marginal likelihood
prob_ik_nonempty_function.R : computes the probability of allocating an individual to a non empty cluster
gamma_causal.R : computes the causal effects at subject-specific level
Gibbs_collapsed_nodags.R : implements the no DAG version of our MCMC algoriithm
Gibbs_collapsed_oracle.R : implements the ORACLE version of our MCMC algorithm
theta_function.R : draws from the posterior of DAG parameters
GIBBS_joint_rcpp.R : contains the collapsed MCMC and retrieves the DAG parameters

The folder data contains the R codes for the the analysis of cardiac side effects induced by anticancer treatments on breast cancer patients.

breast_cancer_reduced.csv : contains the data used in the analysis.
Run_MCMC_reduced.R : implements the MCMC on breast cancer patients and it produces plots for the analysis on the results