v1.1.1

• Separate the generate and preprocess process
• Update pair_nn.* and pair_nn_replica.* for LAMMPS 29Oct2020 or later
• Inconsistent position reading depending on ASE version.
• Sufficient memory allocation for multinary ( > 4) system.
• Fix memory leak

v1.1.0

• Add atomic uncertainty resolution
  • W. Jeong D. Yoo, K. Lee, J. Jung and S. Han, J. Phys. Chem. Lett. 2020, 11, 6090-6096
• Fix memory leak
• Fix atomic uncertainty bug

v1.0.0

• Term fix: epoch → iteration
• Stress training
  • If use_stress true, virial stress in the dataset will be used in training process
• LAMMPS optimization
  • The speed of LAMMPS boosts up around 2 times using memoization algorithm

v0.8.0

Many changes!

• Fixed MPI issues and added MPI support for GDF calculation.
• Fixed compatibility issues.
• Updated documents and examples.
• Added more information to LOG.

and more...

v0.6.0

• Added brace expansion support in str_list.
• Added early stopping feature.
  • if total_epoch < 0, early stop mode activates.
  • Currently, if validation loss is larger than the previous save point, we do not save in the current state.
  • Now, break count is also increased in that point.
  • if break_count > break_max(user can set this in input.yaml), then training is stopped.
  • if current epoch > |total_epoch|, training is also stopped.

v0.5.0

Add new options under neural_network: echeck, fcheck, and stddev.

• If echeck is false, do not check energy RMSE when saving.
• If fcheck is false, do not check force RMSE when saving.
• stddev is standard deviation used to initialize weights.

v0.4.6

• Fixed a bug where regularization is not applied.
• Now the program warns about unidentified options in input file. Unidentified means it is not defined in the default inputs.

v0.4.5

• User-defined optimizer support is added.
• Now elapsed time is averaged over show_interval steps.

v0.4.3

• Changed save mechanism. Now the network is only saved when the validation RMSE is lower than the previous save (RMSE is checked at every show_interval, and save_interval is the minimum interval between saves).