Materials Project formation-energy and band-gap datasets (crystal structures and target values)

Two datasets used in an intrinsic-dimension analysis of materials graph neural networks. Each JSON (gzipped) is a list of records: material_id (real Materials Project id), structure (pymatgen Structure.as_dict(); reconstruct with pymatgen.core.Structure.from_dict), and the target value. MP_eform: formation_energy_per_atom in eV/atom (154,373 entries). MP_bandgap: band_gap in eV (154,879 entries). Load example: import gzip, json; from pymatgen.core import Structure; d = json.load(gzip.open('MP_eform_dataset.json.gz')); s = Structure.from_dict(d[0]['structure']).