This repository contains code and data needed to reporduce the analysis in our paper: Elucidation of Relaxation Dynamics Beyond Equilibrium Through AI-informed X-ray Photon Correlation Spectroscopy by J. P. Horwath, et al, available here.
We have included scripts for building/augmenting a dataset of XPCS two time correlation functions, training a convolutional autoencoder to encode and reproduce XPCS data, and analysis scripts for unsupervised clustering in the latent space.
Most data files needed to run training and analysis are available in the analysis_data directory,
however large files must be downloaded from our storage
server, and files should be placed in
the analysis_data directory.
The AutoEncoder directory contains data augmentation, training, and inference scripts along with pretrained model weights. The XPCSData directory contains the raw data from the rheoXPCS experiment, and the analysis_data directory contains metadata (time and scattering direction), and the encoded XPCS dataset.
All analysis presented in the paper and supplemental information can be reproduced using the Analysis.ipynb notebook, while additional investigation into our choice of clustering algorithm and number of clusters is found in ClusteringEvaluation.
See requirements.txt for python package dependencies. This code has been tested on MacOS Ventura 13.4.1.