Change Log#

v0.4.0 (2022/04/27)#

Added#

  • Add ParameterTransform class to transform parameters into a different space (e.g. log space) @yonatank93

  • Add Weight class to set weight for energy/forces/stress. This is not backward compatible, which changes the signature of the residual function. Previously, in a residual function, the weights are passed in via the data argument, but now, its passed in via an instance of the Weight class. @yonatank93

Fixed#

  • Fix checking cutoff entry @adityakavalur

  • Fix energy_residual_fn and forces_residual_fn to weigh different component

Updated#

  • Change to use precommit GH action to check code format

v0.3.3 (2022/03/25)#

Fixed#

  • Fix neighlist (even after v0.3.2, the problem can still happen). Now neighlist is the same as kimpy

v0.3.2 (2022/03/01)#

Added#

  • Enable params_relation_callback() for KIM model

Fixed#

  • Fix neighbor list segfault due to numerical error for 1D and 2D cases

v0.3.1 (2021/11/20)#

  • add gpu training for NN model; set the gpu parameter of a calculator (e.g. CalculatorTorch(model, gpu=True)) to use it

  • add pyproject.toml, requirements.txt, dependabot.yml to config repo

  • switch to furo doc theme

  • changed: compute grad of energy wrt desc in batch mode (NN calculator)

  • fix: set fingerprints_filename and load descriptor state dict when reuse fingerprints (NN calculator)

v0.3.0 (2021/08/03)#

  • change license to LGPL

  • set default optimizer

  • put kimpy code in try except block

  • add state_dict for descriptors and save it together with model

  • change to use loguru for logging and allows user to set log level

v0.2.2 (2021/05/24)#

  • update to be compatible with kimpy v2.0.0

v0.2.1 (2021/05/24)#

  • update to be compatible with kimpy v2.0.0

  • use entry entry_points to handle command line tool

  • rename utils to devtool

v0.2.0 (2021/01/19)#

  • add type hint for all codes

  • reorganize model and parameters to make it more robust

  • add more docstring for many undocumented class and functions

v0.1.7 (2020/12/20)#

  • add GitHub actions to automatically deploy to PyPI

  • add a simple example to README

v0.1.5 (2020/2/13)#

  • add neighborlist utility, making NN model independent on kimpy

  • add calculator to deal with multiple species for NN model

  • update dropout layer to be compatible with the pytorch 1.3

v0.1.4 (2019/8/24)#

  • add support for the geodesic Levenberg-Marquardt minimization algorithm

  • add command line tool model to inquire available parameters of KIM model

v0.1.3 (2019/8/19)#

  • add RMSE and Fisher information analyzers

  • allow configuration weight for ML models

  • add write optimizer state dictionary for ML models

  • combine functions generate_training_fingerprints() and generate_test_fingerprints() of descriptor to generate_fingerprints() (supporting passing mean and stdev file)

  • rewrite symmetry descriptors to share with KIM driver

v0.1.2 (2019/6/27)#

  • MPI parallelization for physics-based models

  • reorganize machine learning related files

  • various bug fixes

  • API changes * class DataSet renamed to Dataset * class Calculator moved to module calculators from module calculator

v0.1.1 (2019/5/13)#

  • KLIFF available from PyPI now. Using $pip install kliff to install.

  • Use SW model from the KIM website in tutorial.

  • Format code with black.

v0.1.0 (2019/3/29)#

First official release, but API is not guaranteed to be stable.

v0.0.1 (2019/1/1)#

Pre-release.