Preparing training scripts for SchNET, PAIN and ANI

[wiederm]

Description

I am preparing training scripts for SchNET, PaiNN, and ANI. This PR is the first of a series of refactoring PRs.

Changes to NNPs

• NNPs now all inherit from Pytorch Lightning. This avoids keeping track of a Lightning Mixin, separate Lightning classes per NNP and class complexity.
• NNPs now all have:
  • RepresentationLayer (transforms the input)
  • InteractionLayer (performs the NNP compuation) and
  • Readout (performs the scatter add operation to obtain molecular properties)
• NNPs initialize all used modules internally, streamlining the setup and avoiding system setup mistakes while removing some of the flexibility.

Todos

Notable points that this PR has either accomplished or will accomplish.

• [x] training scripts

Status

• [x] Ready to go

[codecov-commenter]

Codecov Report

Merging #75 (cbf1ea9) into main (cfd020b) will decrease coverage by 0.04%. The diff coverage is 94.11%.

Additional details and impacted files