v1.0
JOSS Review
Hoomd-TF has been published as a peer-reviewed article in the
Journal of Open Source Software (JOSS)
New Features
- Added pre-built molecular features
- Added MDAnalysis style selection for defining mapping operators
Enhancements
- Docs can now be built without HTF install
- mol batching performance is much better
- Simplified variable saving
- More example notebooks and reduced file sizes of example trajectories
- Supports dynamic boxes
- Better EDS bias naming
- Prevents accidentally computing forces from positions, instead of nlist
- Added guards against compiler mismatch
- Added sanity tests to prevent unsupported CPU/GPU modes
- Added benchmarking script
- Added check for nlist overflows on GPU
- Added check for mismatch on saving variables/tensors
- Fixed all compiler warnings
- Added Dockerfile for containerized use
Bug Fixes
- Empty tensorboard summaries no longer crash
- Prevented import issues with name clashes between packages and classes