bnRep: A repository of Bayesian networks from the academic literature

Manuele Leonelli
2024-09-28
Abstract:Bayesian networks (BNs) are widely used for modeling complex systems with uncertainty, yet repositories of pre-built BNs remain limited. This paper introduces bnRep, an open-source R package offering a comprehensive collection of documented BNs, facilitating benchmarking, replicability, and education. With over 200 networks from academic publications, bnRep integrates seamlessly with bnlearn and other R packages, providing users with interactive tools for network exploration.
Artificial Intelligence,Physics and Society
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the relatively limited pre - construction and documentation repositories of current Bayesian Networks (BNs). Although Bayesian Networks are widely used in many fields and have a strong ability to model complex systems, the existing BN repositories are small in scale, poorly documented, and not updated in a timely manner, unable to meet the needs of research and practical applications. ### Specific problems include: 1. **Lack of comprehensive BN repositories**: Compared with other machine - learning fields such as deep learning, there are fewer pre - constructed model repositories for Bayesian Networks. For example, TensorFlow Hub and PyTorch Hub provide a large number of pre - trained models, while the relevant resources for Bayesian Networks are relatively scarce. 2. **Limitations of existing BN repositories**: Although several existing BN repositories (such as Bayesian Network Repository, BNMA BN repository, etc.) have a certain number of networks, they are small in scale, not detailed enough in documentation, and not frequently updated. 3. **Promoting reproducibility and benchmarking**: The lack of a comprehensive BN repository makes it difficult for researchers to conduct effective benchmarking and reproduce results, affecting the reliability and reproducibility of research. 4. **Insufficient educational and learning tools**: For students and practitioners, there is a lack of a centralized and easily accessible resource to learn and explore the applications of Bayesian Networks. ### Solutions: To this end, the author has developed an open - source R package named `bnRep`, aiming to provide a comprehensive repository containing more than 200 Bayesian Networks from academic literature. The features of `bnRep` include: - **Rich network resources**: It contains more than 200 Bayesian Networks, and each network is accompanied by detailed documentation. - **Seamless integration with existing tools**: It is compatible with commonly used R packages such as `bnlearn`, and supports three types of Bayesian Networks: discrete, Gaussian, and conditional linear Gaussian. - **Interactive exploration tools**: It provides a Shiny application, and users can easily browse and filter the network database through a browser. - **Promoting research and education**: It provides resources for researchers to conduct benchmarking and reproduce results, and is also an excellent learning tool. Through these measures, `bnRep` not only fills the gap in Bayesian Network resources but also lays the foundation for future research and interdisciplinary applications.