Processed Data for Results in Crystal Hypergraph Neural Networks

materials_hypergraph_data contains the subdirectories:

proc_materials_project - processed data for Materials Project target sets (band_gap, formation_energy, metalicity)

proc_log_k                    - processed data for MatBench bulk moduli target set

proc_log_g                    - processed data for MatBench shear moduli target set

proc_dielectric              - processed data for MatBench dielectric constants target set

proc_perov                    - processed data for MatBench perovskite formation energy target set

proc_phonon                - processed data for MatBench highest frequency phonon peak target set

Each processed dataset contains a data file for every material with hyperedge indices (for motifs and triplets), edge indices (for bonds), edge features (Gaussian expansion of interatomic distance as edge_attr), and hyperedge features (Gaussian expansion of angle for triplets and local structure order parameters and continuous symmetry measures for motifs) saved as attributes of a pytorch geometric Graph object. Target values are saved as 'y' attributes for MatBench target sets. For materials project data with multiple targets for each material, targets are saved as separate 'band_gap', 'form_en', and 'metalicity' attributes in the same Graph object.