Bayesian Hidden Physics Models

Code accompanying Atkinson et al., Discovery of Physics and Characterization of Microstructure from Data with Bayesian Hidden Physics Models

Getting set up

Data

Download the required data for the examples here. Unzip so that you have the following files relative to this README:

• ultrasound_data/Hsample SAW 5MHz n2/wvf.mat
• ultrasound_data/30Jan15 Nist crack 240x240 12x12mm avg20 5MHz 0deg grips/wvf.mat

Configuring your environment

Experiments were run on a desktop running Ubuntu 18.04 with Python 3.7.3 with dependencies enumerated in requirements-explicit.txt. A less verbose set of dependencies that should still work is given in requirements.txt.

Getting set up should be as simple as pip install with the right flags (i.e. -r and -f for JAX). However, we've provided a convenience script; just do

. configure.bash

and you should be good. Note that you may need to alter some parameters at the top of the script (e.g. CUDA version) to your specific setup.

Running examples

A sequence of bash scripts are provided in examples to recreate the results reported in the paper. Core functionality is packaged in the bhpm module.