NeSy-for-graph-data

Code for the paper: "A Neuro-Symbolic Approach for Probabilistic Reasoning on Graph Data"

Repository Structure

water-hawqs-github/

• Contains model definitions, training scripts, and dataset creation tools for the HAWQS dataset.
• Precomputed result graphs are stored as pickle files in water-hawqs-github/data.
• Note: Raw HAWQS/SWAT data is omitted to reduce repository size. The /scenarios folder contains only a single example scenario.

hetero-hom-experiments/

• Contains experiments with the Ising model, including dataset generation, model training, and data export for Primula.
• Includes additional experiments using the GMNN model.

GNN-RBN-compile/

• Slight adaptation from the experiments of "Generalized Reasoning with Graph Neural Networks by Relational Bayesian Network Encodings"
• Add ACR-GNN compilation support for categorical variables and for parameter learning in Primula
• .rbn file for the PyTorch-GNN interface

Notebooks

• Both main directories include Python notebooks for data exploration and preprocessing.

Primula Integration

• Experiments involving Relational Bayesian Networks (RBNs) use Primula-3 for integration.

Folder Overview

• water-hawqs-github/: Main codebase for HAWQS-related experiments and data.
• water-hawqs-github/data/: Precomputed results (pickle files) for quick access.
• water-hawqs-github/scenarios/: Example scenario for HAWQS (raw data not included).
• hetero-hom-experiments/: Scripts and data for Ising model and GMNN experiments.
• Notebooks/: Jupyter notebooks for analysis and preprocessing (located in both main folders).