schnetkit
some tooling for schnetpack
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
asecalculator forSchNetmodels 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.