CCDr Algorithm for Learning Sparse Gaussian Bayesian Networks

Implementation of the CCDr (Concave penalized Coordinate Descent with reparametrization) structure learning algorithm as described in Aragam and Zhou (2015) < https://www.jmlr.org/papers/v16/aragam15a.html>. This is a fast, score-based method for learning Bayesian networks that uses sparse regularization and block-cyclic coordinate descent.

ccdrAlgorithm

ccdrAlgorithm implements the CCDr structure learning algorithm described in [1-2]. This algorithm estimates the structure of a Bayesian network from mixed observational and experimental data using penalized maximum likelihood based on L1 or concave (MCP) regularization.

Presently, this package implements the main algorithm and provides a method to simulate data from a Gaussian Bayesian network. To simulate random networks, it is recommended to use the sparsebnUtils package. Other packages for simulating DAGs and observational data include bnlearn, pcalg, and igraph.

Overview

The main method is ccdr.run, which runs the CCDr structure learning algorithm as described in [1-2]. For simulating data from a Gaussian Bayesian network, the package provides the method generate_mvn_data. This method can simulate observational data or experimental data with interventions (or combinations of both).

Installation

You can install:

• the latest CRAN version with

  install.packages("ccdrAlgorithm")
  

• the latest development version from GitHub with

  devtools::install_github(c("itsrainingdata/sparsebnUtils/dev", "itsrainingdata/ccdrAlgorithm/dev"))
  

References

[1] Aragam, B. and Zhou, Q. (2015). Concave penalized estimation of sparse Gaussian Bayesian networks. The Journal of Machine Learning Research. 16(Nov):2273−2328.

[2] Zhang, D. (2016). Concave Penalized Estimation of Causal Gaussian Networks with Intervention. Master’s thesis, UCLA.

[3] Fu, F. and Zhou, Q. (2013). Learning sparse causal Gaussian networks with experimental intervention: Regularization and coordinate descent. Journal of the American Statistical Association, 108: 288-300.