Welcome to hippynn’s documentation!
We hope you enjoy your stay.
What is hippynn?
hippynn is a python library for machine learning on atomistic systems
using pytorch.
We aim to provide high-performance modular design so that different
components can be re-used, extended, or added to. You can find more information
about overall library features at the hippynn Features page.
The development home is located at the github github repository, which also contains many example files.
Additionally, the user guide aims to describe abstract
aspects of the library, while the
examples documentation section aims to show
more concretely how to perform tasks with hippynn. Finally, the
api documentation contains a comprehensive
listing of the library components and their documentation.
The main components of hippynn are constructing models, loading databases, training the models to those databases, making predictions on new databases, and interfacing with other atomistic codes for operations such as molecular dynamics. In particular, we provide interfaces to ASE (prediction), PYSEQM (training/prediction), and LAMMPS (prediction). hippynn is also used within ALF for generating machine learned potentials along with their training data completely from scratch.
Multiple database formats for training data are supported, including Numpy arrays, ASE-compatible formats, FitSNAP JSON format, and ANI HDF5 files.
hippynn includes many tools, such as an ASE calculator,
a LAMMPS MLIAP interface,
batched prediction and batched geometry optimization,
automatic ensemble creation,
restarting training from checkpoints,
sample-weighted loss functions,
distributed training with pytorch lightning,
and more.
hippynn is highly modular, and if you are a model developer, interfacing your
pytorch model into the hippynn node/graph system will make it simple and easy for users
to build models of energy, charge, bond order, excited state energies, and more.
Getting Started:
User Guide:
Examples:
API Documentation: