What is embed_files in Custom Dataset

[julien-blanchon]

torchmdnet.dataset.custom.Custom take 3 arguments: coordglob, embedglob, energyglob.

For each file:

• Coord is a numpy array of lenght N with N the number of atom in the molecule
• What is embed ? My first guess was a numpy array with the Z of each atom, but this don't match the dimension
• What is energy ? Energy array seems to match the dimension of the coord array

[RaulPPelaez]

Maybe this test will help clarify: https://github.com/torchmd/torchmd-net/blob/e20876ff79fe952db25e5fc55cc051a919ef218f/tests/test_datasets.py#L13-L45

• coord is an array of shape (n_molecules, n_atoms, 3)
• embed (could be z or whatever you want to use for the embedding) is an array of shape (n_atoms,), note that the class refers to it as "z" anyhow.
• energy is an array of shape (n_molecules, 1)

[RaulPPelaez]

I am closing this, please feel free to reopen if you have further questions about it!

[julien-blanchon]

Yes this was clear thanks !