- Disable progress bar for readability.
- Disable notifications.
- Enable isolated atom calculations
- Security fix of tensorflow
- Docstring fix @a-ws-m
- Minor format fixes
- Revert to use tf.gather with warnings for speed concerns
- Add swish activation function
- Remove training warning
- Update tutorial
- Stability improvements
- Include data files in wheel
- Add pre-trained multi-fidelity models
- Add utils for disordered site representation
- Add more multi-fidelity plots
- Revising plotting of multifidelity work
- Release for minting DOI
- Bug fix
- Add multi-fidelity model examples
- Add sample weights for model training
- Add default optimizer gradient norm clip
- Bug fix of megnet descriptors
- Update the model training mechanism and move Gaussian expansion to tensorflow, training speed up 100%
- minor fix to include more linear readout option
- add data type control
- refactor local_env
- Code refactor and reformat
- Fix tensorflow and numpy type compatibility issues
- Update to tensorflow.keras API instead of using keras
- Download mvl_models from figshare if not present
- Add mvl_models in wheel release file
- Bug fix and version correction
- Fix bug brought by migrating to tensorflow 2.0
- New elasticity models trained on 2019 MP data base
- Add meg command line tools
- Add mypy typing hint for non-tensorflow codes
- Update keras to 2.3.1 to fix thread-safety issues
- New find_points_in_spheres algorithm in pymatgen for graph construction
- Tensorflow 2.0 version.
- Change
convertortoconverterin all model APIs - Improve
ReduceLRUponNancallback function - @WardLT major contributions to the
MolecularGraphclass - Add serialization methods for
local_envclasses - delete
data/mp.py
- GraphModel and MEGNetModel now supports a metadata tag, which is included in the JSON. (suggestion of @mhorton).
- Misc bug fixes for edge cases as well as improved error messages for mismatches in inputs.
- Implement the option for a scaler in models, which is used in efermi models at the moment but also can be helpful for extensive quantities.
- Minor fixes to setup.py and licenses.
- Proper fix to release on PyPi.
- Major refactoring to conform to OOP principles. Note that the changes are not backwards compatible, but many things are a lot simpler. We do not expect much disruption to existing users.
- Added pre-trained models developed in our work for users who wish to simply use them for prediction.
- Major improvements to README and documentation.
- Bug fix for dimension problem when only one atom in structure
- Initial release