v0.5.0

v0.5.0 Release

notable changes:

• New learning rate schedules
• New batch-evaluationb method for the ase calculator
• h5md dataset loading
• new dataset optimized for differently sized structures
• new Bessel-like basis functions
• shallow ensembles for uncertainty estimation and NLL, CRPS loss functions
• added feature model
• switched linting to ruff
• force features for BAL

What's Changed

• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #266
• add repository link by @PythonFZ in #267
• EMA, SAM update, cyclic LR schedule by @M-R-Schaefer in #269
• train_config.py fix default no. contractions by @Tetracarbonylnickel in #272
• NVE bug fix by @Tetracarbonylnickel in #276
• Add MD restart support and fix config loading by @PythonFZ in #271
• bump znh5md by @PythonFZ in #278
• implemented apax batch evaluation node by @M-R-Schaefer in #280
• H5MD data loading by @M-R-Schaefer in #282
• Per Batch Padded Dataset by @M-R-Schaefer in #281
• allow for non NTK linear layers in model by @M-R-Schaefer in #283
• New Basis function interface by @M-R-Schaefer in #284
• Mlflow fix by @M-R-Schaefer in #286
• Node fix by @M-R-Schaefer in #289
• fixed memory leak in pbp ds by @M-R-Schaefer in #290
• Shallow ensembles by @M-R-Schaefer in #275
• refactored distance computation, added feature model by @M-R-Schaefer in #285
• use znh5md v0.3 by @PythonFZ in #292
• replace isort, flake8 and black with ruff by @PythonFZ in #294
• Collective changes from my absence by @M-R-Schaefer in #293
• bump project version, pin orbax and flax versions by @M-R-Schaefer in #296
• v0.5.0 by @M-R-Schaefer in #297

Full Changelog: v0.4.0...v.0.5.0