CodietPGM

This repository represents a toolkit for structure learning of DBNs in the CoDiet [1] project. Currently, this repository is under development. The goal is to finally extend this repository into a toolkit that would complement pgmpy library [2] in the case of Dynamic Bayesian Networks.

The repository contains several bundles of code. There are structured as follows:

• The io folder contains tools for loading and processing data. The organization of the data is focused on the types of data that will arise in the CoDiet project. The Data class contains a collection of Samples, which will represent individual patients. Each Sample has a set of static features, and a set of dynamic features that are stored in a data frame. For each of the variables, annotations are available, so that a relevant set of variables can be selected for an experiment.
• Once the data are loaded, the user can use one of the learners to learn the structure and parameters of the model. There are several learners available. For some, external libraries are used, as in the case DyNoTearsDBN, some are related to publication [3], as MILPDBN.
• The structure package contains a representation of the learned model. The model consists of instances of Node, which represent individual variables. Then, each node is attached transitions that have a probability distribution over input_nodes. See file makeafewDBNs.py in tests for example of structure creation.
• In utils, you can find helper code.
• The evaluation folder (under development) will contain tools to compare different results obtained by learners.
• Folder dag_gflownet contians a snapshot of the GFlowNet package [6]. As this package cannot be installed using pip (yet), we decided to include its copy in here.
• In gfn directory, you can find code for using GFlowNet.
• Folder R_codes contains R code that can be used to learn using BNStruct [4] and MCMC BiDAG package [5].

[1] https://www.codiet.eu/

[2] https://pgmpy.org/

[3] https://arxiv.org/abs/2406.17585

[4] https://cran.r-project.org/web/packages/bnstruct/index.html

[5] https://cran.r-project.org/web/packages/BiDAG/index.html

[6] https://github.com/alexhernandezgarcia/gflownet