Data Archive for "Probing the effects of broken symmetries in machine learning"

This collect the input files, scripts, and data for the publication:

Title: Probing the effects of broken symmetries in machine learning
Authors: Marcel F. Langer, Sergey N. Pozdnyakov, and Michele Ceriotti
in Mach. Learn.: Sci. Technol. 5 04LT01
DOI: 10.1088/2632-2153/ad86a0
Preprint: arxiv:2406.17747 (2024)

The data in this archive is stored in the Materials Cloud Archive under DOI:10.24435/materialscloud:kz-3b, and mirrored at github.com/sirmarcel/eqt-archive.

---

The following items are contained within:

• equivator-figs.ipynb: Notebook to generate all figures in the publication (also contains some additional analysis used in the manuscript)
• figures/: Figures from the publication
• data/: The raw data used to generate the figures
• md_templates/: Input files for (gas, liquid) MD runs
• scripts/: Various Python scripts to process MD runs into the data used for publication
• ice/: Experiment for water ice
• model/: Contains the PET model used for this work

Additional README.md files are supplied to give more detailed explanations.

The main tools used in this work are the PET model, which is implemented in this repository: https://github.com/spozdn/pet, and the I-Pi code, obtained from https://github.com/i-pi/i-pi/ (@99fcf1). Experiments used a custom branch of PET, which can be found under https://github.com/lab-cosmo/pet/tree/neighbors_convert_cpp (@e2631cc) -- the functionality will be merged into main PET eventually.

---

To download the repository:

• On GitHub, you simply can run git clone, provided that git-lfs is enabled.
• On Materials Cloud Archive, the entire contents listed above are provided as a .zip file.

The files in this repository are licensed under the Creative Commons Attribution 4.0 International license.