About The Project

The D&C order pruner is an exact structure learning algorithm for Bayesian networks, operating on the space of topological orders and using a divide and conquer technique.

Getting Started

Clone the repository and make

git clone https://github.com/felixleopoldo/dncpruner.git

Easiest way to run the algorithms is to build a Docker image from the Docker file and run the algorithm in a container.

Docker

    $ docker build . -t dnc

Run the container and mount the current directory, with the repo

    $ docker run -w /mnt -v $(pwd):/mnt -it dnc

Inside the container run

make

Now follow the instructions below

Native

Requirements

• Boost 1.82.0 C++ library

R packages

• RInside
• Rcpp
• Jmisc
• argparser
• igraph
• BiDAG 2.0.0

For plotting:

• ggplot2
• dplyr
• latex2exp
• patchwork

Installation

After cloning th repo, type make

make

(back to top)

Usage

To run the algorithm with the data file data/p20n300gaussdata.csv type

./dncpruner --filename data/asiadata.csv --scoretype bge --am 0.1 --aw NULL --output_csv dag_adjmat.csv

The estimate dag is then as an adjacency matrix in the CSV file dag_adjmat.csv. If M_ij=1 in the adjacency matrix, then there is an edge i->j in the DAG.

Benchmarks

To generate benchmarks (with the setting specified in R/benchmark_dncpruner.R) type

Rscript R/benchmark_dncpruner.R --output_dir results --filename res.csv --seeds_from 1 --seeds_to 5

This produces joined results in the file res.csv which can be analysed by typing

source('R/plotting.R')

(back to top)

License

Distributed under the Apache 2.0 License. See LICENSE.txt for more information.

(back to top)