xnn.common.data.hub.qm7#

QM7 dataset builder.

QM7 (Blum & Reymond 2009; Rupp et al., PRL 108, 058301, 2012) holds 7165 organic molecules of up to 23 atoms (H, C, N, O, S) from GDB-13 with their PBE0 atomization energies, the benchmark of the spherical CNN of Cohen et al. (ICLR 2018, Sec. 5.4). The authors provide a stratified five-fold split.

Data: http://quantum-machine.org/datasets/ (qm7.mat).

Notes

  • Upstream positions are in Bohr and energies in kcal/mol; by default both are converted to Angstrom and eV (units="eV").

  • The file is a MATLAB .mat; reading it needs scipy.

Classes

QM7Builder()

Builder for QM7 (7165 molecules, atomization energies, five folds).
Show Source