schnetkit
schnetpack
some tooling for EXPERIMENTAL/IN DEVELOPMENT DO NOT USE
This is an early draft of some infrastructure built around schnetpack
. In particular, it allows:
- Saving/loading models from disk
- Initialising models from a
.yaml
-based description of the architecture - Efficient
ase
calculator forSchNet
models with support for a larger effective cutoff (i.e. avoiding to recompute the neighborlists on every step)
As a result of this, the schnetkit
calculator is compatible with FHI-vibes
. (Currently, this functionality is untested.)
Future versions will support:
- Restarteable training with automatic re-submission in
slurm
- Training configurable with input files
- Efficient dataset class that pre-computes everything
This is a part of the infrastructure for the gknet
project, and will be completed as the project nears publication.