All notable changes to this project will be documented in this file.
- Normalization procedure on-the-fly via nnp-train, keyword
normalize_data_set
. - New normalization scheme based on force predictions.
- New keyword
force_energy_ratio
for nnp-train. - Additional neighbor statistics output from nnp-scaling.
- New tool nnp-checkdw (for testing purposes).
- New CI via Github Actions.
- Use self-compiled doxygen for API docs, use version 1.9.4.
- Fixed missing atomic energy offsets for LAMMPS interface.
- Fixed NaNs from nnp-dist when running mixed data sets.
- Fixed bugs in pynnp examples.
- Changed build behavior for LAMMPS interface to conform with PR (lammps/lammps#2626)
- Fixed bug in nnp-scaling which could cause segfaults (MPI only).
- New tool to check numeric vs. analytic forces
- Fixed critical bug for large cutoff values (large compared to lattice vectors)
- Now compatible with latest LAMMPS stable release 29Oct2020
- Improved EW warnings in LAMMPS
- Citation header in screen output
- Restructured training to prepare for multi-network HDNNPs (e.g. 4G-HDNNP)
- New training "properties" can be added more easily now
- Fixed bugs in makefiles
- Fixed bug when creating scaling data
- Polynomial symmetry functions (Martin P. Bircher and Andreas Singraber)
- Better symmetry function cache mechanism
- Documentation updates (API docs in Sphinx via Breathe)
- Tool nnp-scaling puts out rejected configurations into a separate file.
- Bugfix in CabanaMD interface: SF group initialization
- Fix bug in makefile which sometimes causes a failed build due to parallel execution
- Training library (Multi-stream Kalman filter training).
- Tools for HDNNP data set handling, etc.
- Python interface (basic functionality).
- Sphinx documentation (+ Doxygen API reference).
- CabanaMD interface (by Saaketh Desai and Sam Reeve)
- License change from MPL 2.0 to GPL v3 or later.
- Core library (NN, symmetry functions, ...).
- LAMMPS interface.
- Documentation (in parts) via doxygen.