EBFormer Datasets

The datasets used in EBFormer are included in this upload. A brief description is included below: Dataset 0.3: Contains all data used in the interpolation task. Subsets of 1k, 2k, and 4k are used for the learning curve. Dataset 1.3: Contains all data used in the zero-shot out-of-distribution task. Dataset 2.4: Contains all data used in the transfer learning task presented in the supplementary information. Each folder includes the specific test, train, and validation sets for the results in the EBFormer paper. All datasets are formatted as extxyz files, with key 'ldos' and 'lJx' indicating locally-projected DOS/cDOS. '_per_atom' quantities contain the mean-pooled local values. Positions are in Angstroms, DOS in 1/eV, and cDOS in 1/eV 1/cm Hz.