search

gasteigerjo / dimenet

DimeNet and DimeNet++ models, as proposed in "Directional Message Passing for Molecular Graphs" (ICLR 2020) and "Fast and Uncertainty-Aware Directional Message Passing for Non-Equilibrium Molecules" (NeurIPS-W 2020)
https://www.daml.in.tum.de/dimenet
Other
292 stars 61 forks source link

Coll dataset units #29

Closed spozdn closed 1 year ago

spozdn commented 1 year ago

Dear authors, Could you confirm that the units of the COLL dataset available here https://figshare.com/articles/dataset/COLL_Dataset_v1_2/13289165/1 are eV and eV/A for energies and forces, respectively (some readme would be really helpful :) )?

Also, could you confirm that the errors in forces reported in DimeNet and GemNet papers are per a single xyz component?

gasteigerjo commented 1 year ago

Hi Sergey!

Thank you for your interest!

The units in COLL are indeed eV and eV/A. Note that the atomization energy is the relevant one for prediction tasks.

The errors in DimNet and GemNet are mean absolute errors (MAEs), as defined e.g. in the second term of Equation 2 in the DimeNet (not DimeNet++) paper. You can also see this in the code here.

spozdn commented 1 year ago

Thanks!