Training on the Dataset without PBC

[almakhmudov]

Dear community,

I was wondering if it would be possible to train NequIP on a dataset without lattice parameters and PBC set to false? Consider I have a dataset with a chemical reaction involving one molecule only.

I would greatly appreciate any help!

Best,
Albert

[cw-tan]

Hi @almakhmudov ,

Thanks for your interest in the code. It should be possible, just make sure that is reflected in the construction of your extended xyz file if using that format. See https://wiki.fysik.dtu.dk/ase/ase/io/formatoptions.html#extxyz. Hopefully you don't need to look at the following link, but it doesn't work out perfectly, you can debug through this function to check if the PBC, etc is passed correctly:

nequip/nequip/data/AtomicData.py

Line 390 in 1e150cd

def from_ase(

[almakhmudov]

Thanks a lot for the reply. Don't you mind if I ask a follow-up question. I was wondering how could I run nequip-evaluate on a completely different dataset? I would imagine that I'd have to use the same config file and perhaps change a path to a dataset. At the same time what should I do with 'n_train' and 'n_val' keywords? Not using them in a config file would be the easiest solution. Either way looking forward to your response.