Beyond Training Data How Elemental Features Enhance ML-Based Formation Energy Predictions
Published: 12 August 2025| Version 2 | DOI: 10.17632/n3cwj2hb7w.2
Contributors:
Hamed Mahdavi, , Description
This repository contains the code and data for the project "Beyond Training Data How Elemental Features Enhance ML-Based Formation Energy Predictions".
Files
Steps to reproduce
Copy the data folder into the code folder and use the commands provided in run-ood-experiments.sh to reproduce the results. The outputs will be logged via TensorBoard. The ood-list experiment type re-runs the experiments with Out-of-Distribution (OoD) training data, using the set of excluded elements specified by the excluded_elements_list argument in code/evaluate.py. To repeat experiments where the training data remains complete but the validation set includes only compounds containing one of the excluded elements, use the experiment type ood-list-train-iid-scaled
Institutions
University of Wisconsin Madison, Pennsylvania State University
Categories
Materials Science, Crystal, Material Property Database