DynamicalGraphModel project

Welcome! This project provides Python code for generating and solving systems of equations describing a specified compartmental model on a specified graph (network).

This project is used in a publication¹ that assesses more broadly this approach to compartmental modelling on graphs. There is documentation throughout the Python code, but to get started you should read the next section.

Getting started

• Go to equation.__init__.py to see an example of how equations can be generated (and printed), initial conditions can be specified and equations solved for these conditions.
  • Models can be specified with a graph (we use networkx) and a compartmental model (we use a simplified version of CModel).
  • Compartmental models can be as simple as SIR (model_params.CModel contains a getter method for the usual form of this)
  • Graphs can be generated using networkx methods for simple graph classes or user-specified
  • Initial conditions can be generated using helper methods - see example usage in the __init__ file
• If you'd like to compare to a Monte Carlo simulation, go to monte_carlo.equation_MC_comparison.py for examples used in the associated publication

Large-scale experiments

• Use PYTHONPATH=src python -m experiments.run_experiments --help to discover the CLI for running larger runtimes, including 100-vertex graphs, multiple graph families, and long timeouts. Experiments can be described ad-hoc via flags or via a JSON file (see experiments/config.py).
• If --config is omitted the runner automatically consumes src/experiments/default_config.json, which spans Erdős–Rényi, Barabási–Albert, Watts–Strogatz, random regular, and random geometric graphs at 100 vertices with extended timeouts.
• Results are appended to data/experiment_results.csv by default; override with --output.

Remote deployment on fatanodes

• Prepare a JSON configuration with the experiments to run (e.g. experiments.json).
• Stage and launch the code on the remote machines with python -m deployment.remote_launcher --gateway <gateway-host> --local-config experiments.json.
• The launcher syncs the repository to ~/DynamicalGraphModel on the head node, uploads the JSON config, and starts partitioned runs across the compute nodes via nohup, writing logs to ~/DynamicalGraphModel/logs and results to ~/DynamicalGraphModel/data/remote_results.csv.
• Pass --dry-run to inspect the exact SSH/rsync commands without executing them, and adjust --python or --compute-nodes if a custom environment is required.

---

Footnotes

1. Ethan Hunter, Jessica Enright, Alice Miller, Feasibility assessments of a dynamical approach to compartmental modelling on graphs: Scaling limits and performance analysis, Theoretical Computer Science, Volume 980 (2023), doi.org/10.1016/j.tcs.2023.114247 ↩