Version 0.4.3

What's Changed

Bug Fixes 🐛

• Fix installing ptemcee by @mjwen in #144

Full Changelog: v0.4.2...v0.4.3

Version 0.4.2

What's Changed

Enhancements 🛠

• Refactor test by @mjwen in #125
• Update the ptemcee dependency by @yonatank93 in #137
• Update GH actions to use latest conda-forge kim-api and test on macOS by @mjwen in #143

Documentation 📖

• Recreate docs building codes by @mjwen in #129

Other Changes

• Fix neighbor list bug by @mjwen in #90
• Fix _WrapperCalculator by @mjwen in #95
• Remove requirements.txt, add info in setup.py by @mjwen in #108
• Add multiple species support of LJ by @mjwen in #112
• Update CI to fix cmake version by @mjwen in #117
• WIP: Implement bootstrap by @yonatank93 in #107

Full Changelog: v0.4.1...v0.4.2

Version 0.4.1

Added

• Uncertainty quantification via MCMC (@yonatank93). New tutorial and explanation of the functionality provided in the doc. (@yonatank93)
• Issue and PR template

Fixed

• Linear regression model parameter shape (#75)
• NN multispecies calculator to use parameters of all models (#71)

Updated

• Documentation on installing KLIFF and dependences

Version 0.4.0

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 pre-commit GH action to check code format

Version 0.3.3

Fixed

• Even after v0.3.2, the problem in #37 for 1D can still happen. Now fixed and the neighbor list is the same as that in kimpy.

Version 0.3.2

Added

• Enable params_relation_callback() for KIM model

Fixed

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

Version 0.3.1

• 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
• change to use the furo theme for docs
• change to compute grad of energy wrt desc in batch mode (NN calculator)
• fix to set fingerprints_filename and load descriptor state dict when reusing fingerprints (NN calculator)

Version 0.3.0

• 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 allow user to set log level

Version 0.2.2

• Fix pybind11 includes

Version 0.2.1

• update to be compatible with kimpy v2.0.0
• use entry entry_points to handle command line tool
• rename utils to devtool