Wasserstein distance prior impact assessment for ODE models

Introduction

This repository contains supporting code for the pre-print:

Mingo, D. N., Hale, J. S. and Ley, C.,: Bayesian prior impact assessment for dynamical systems described by ordinary differential equations.

TODO: Add link to preprint

The code is archived at:

Mingo, D. N. and Hale, J. S.: Wasserstein distance prior impact assessment for ODE models, https://doi.org/10.5281/zenodo.11553775, 2024.

The code in this repository is licensed under the GNU Lesser General Public License version 3 or later, see COPYING and COPYING.LESSER.

Examples

Lotka-Volterra

Scripts to reproduce the results for the Lotka-Volterra example are contained in examples/lotka_volterra.

In order, run:

1. wasser_exlot.py
2. wasser_dist_prior.py
3. wasser_dist_ex.py

The scripts lotka_priors_ppc.py and pairplot.py can be executed independently.

SEIR

Scripts to reproduce the results for the SEIR example are contained in examples/SEIR.

In order, run:

1. All files in prior_samples/.
2. run_seirpost.sh executes wasser_seir.py and saves the samples.
3. wasser_seir.py
4. wd_mar_ex.py

Prior samples (SEIR)

The subfolder examples/SEIR/prior_samples contains scripts for sampling from prior distributions. Running these scripts on an HPC is preferable as they take longer, or use the script sample_batches.py for batch sampling to reduce execution time.

Additional scripts

The script example_diagnostics.py demonstrates how to perform Geweke diagnostics to check for model convergence. The script requires uploading posterior samples and specifying the model (SEIR or Lotka-Volterra).