v2.0.0

Notes

This is the first release of SchNetPack 2.0 which uses the Hydra configuration framework, Pytorch Lightning and a new indexing scheme.
It also includes an improved data pipeline, modules for equivariant neural networks and a PyTorch implementation of molecular dynamics.

What's Changed

• Inital commit for v1 rewrite by @ktschuett in #267
• center transformations by @Stefaanhess in #271
• Mg/torch env by @mgastegger in #274
• Kts/qm9datamodule by @ktschuett in #275
• Training script driven by Hydra+Lightning by @ktschuett in #277
• Implement training of potential energy surfaces by @ktschuett in #278
• Add split file by @ktschuett in #282
• PaiNN representation by @ktschuett in #284
• Postprocessors and TorchScript by @ktschuett in #285
• Mg/symfuncs by @mgastegger in #287
• Initial API docs by @ktschuett in #293
• Fix problem with transforms in new lightning version by @ktschuett in #294
• stress and custom experiment by @Stefaanhess in #292
• Update docs (and fix postprocess bug) by @ktschuett in #295
• Mg/calculators by @mgastegger in #296
• Refactor models by @ktschuett in #297
• fix small bug by @Stefaanhess in #298
• Sh/datasets by @Stefaanhess in #300
• Proposal for MultiPropertyModel by @Stefaanhess in #299
• add epsilon to painn by @Stefaanhess in #301
• Some minor updates by @ktschuett in #305
• Fix API docs by @ktschuett in #306
• Unify model classes & refactor configs by @ktschuett in #309
• Fix ModelOutput package by @ktschuett in #310
• add bessel representation for future painn usage by @Divide-By-0 in #311
• Update examples & add data workdir by @ktschuett in #314
• Dipole moment & polarizability by @ktschuett in #316
• Dynamics caching neighborlist by @ktschuett in #320
• Dev by @jnsLs in #322
• Add long range cutoff by @ktschuett in #325
• Mg/md by @mgastegger in #315
• Refactor configs and add predict script by @ktschuett in #334
• Fix install bug by @ktschuett in #335
• Add automatic position derivatives by @ktschuett in #341
• Update QM9 tutorial by @ktschuett in #347
• ZBL Potential, Electrostatics (+Ewald summation) and stress tensor fixes by @mgastegger in #349
• Fixes to some MD desfaults and torchscript issues by @mgastegger in #350
• Disable automatic use of torchscript in MD calculators by @mgastegger in #351
• Response properties and field representations by @mgastegger in #339
• Fixed derivative graph settings for basic response properties by @mgastegger in #352
• Fixed sign for shift type cutoff by @mgastegger in #353
• Refactor AtomisticModel by @ktschuett in #354
• Fix bug when using ddp with set run.id by @ktschuett in #359
• Fix mixing bias by @ktschuett in #365
• Fix mixing residual by @ktschuett in #366
• fixed aggregation_mode bug for avg pooling by @jnsLs in #361
• Sh/ep device by @Stefaanhess in #360
• Fix AtomsDataSubset for use inside ConcatAtomsData by @chgaul in #357
• Add learning rate warmup & SGD config by @ktschuett in #370
• Fixed creation of subset in BaseAtomsData by @NiklasGebauer in #369
• Fix DDP training by @ktschuett in #372
• Updated for new yaml behavior by @mgastegger in #375
• Nwag/comment-update by @NiklasGebauer in #374
• Fix OMDB _convert dataset preparation by @bartolsthoorn in #373
• Fix pin_memory and some deprecation warnings by @ktschuett in #378
• Update tutorials and filter outputs by @ktschuett in #377
• Add conversion script for old datasets by @ktschuett in #380
• Fixed bug in map_properties by @mgastegger in #381
• Fix testing bc lightning API changed by @ktschuett in #382
• Updated MD docstrings and tutorial by @mgastegger in #383
• Wrapping of atom positions under PBC by @mgastegger in #385
• Fixed energy logging for multiple molecules by @mgastegger in #386
• Some refactoring and cleanup by @ktschuett in #387
• Retrained ethanol model for new postprocessing convention by @mgastegger in #388
• Added on_step_finalize in MD simulation hooks by @mgastegger in #390
• Updated weight init in FieldSchNet representation by @mgastegger in #391
• Added tmpdir functionality for MDs, fixed calculator bug by @mgastegger in #395
• Improved config loading for MD by @mgastegger in #396
• Fixed criterion for recomputing MD neighborlists by @mgastegger in #398
• Fixed using subset in ASEAtomsData.iter_properties by @NiklasGebauer in #393
• update deprecated code to new torch version by @Stefaanhess in #399
• Fix strain input module by @ktschuett in #403
• Add resolver for tmp directory by @NiklasGebauer in #406
• Fix #401 by @ktschuett in #404
• Added tempfile import for custom tmpdir resolver by @mgastegger in #408
• Added routine to NeighborListMD to properly filter out pairs due to the buffer region by @mgastegger in #409
• Removed n_out argument from DipoleMoment and Polarizability layer docstrings by @mgastegger in #411
• Fix PyTorch Lightning deprecations by @ktschuett in #414
• Consider only a selection of atomic forces in training, validation and testing, and ASE neighborlist with skin implemented by @jnsLs in #405
• Added a few classes util for structure relaxations (in particular MOMONANO) by @jnsLs in #415
• New datasets and fixes by @ktschuett in #417
• Fixed inverted grad context for calculator by @mgastegger in #419
• Update to ASE interface and SchNetPack ASE calculator by @mgastegger in #418
• Added neighborlist from MatScipy package by @mgastegger in #421
• Upgrade to Hydra 1.2 by @ktschuett in #420
• Upgrade Hydra configs by @ktschuett in #424
• Fixed issue with merging loaded configs by @mgastegger in #423
• Fix charge correction in dipole moment by @ktschuett in #425
• Merge SchNetPack2 into master branch by @ktschuett in #426
• additional inputs are now an attribute of nbh list postprocessing cla… by @jnsLs in #427
• Fix error in AddOffsets by @NiklasGebauer in #432
• Fixed a problem with hyperparameter logging by @mgastegger in #429
• Remove some unneccesary code and simplify worker defaults by @ktschuett in #434
• Bump protobuf from 3.20.1 to 3.20.2 in /docs by @dependabot in #437
• Bump protobuf from 3.20.1 to 3.20.2 by @dependabot in #436
• Add infos for atomref by @ktschuett in #439
• implemented torch lbfgs optimizer that allows for relaxation of multiple structures in parallel by @jnsLs in #438
• Add workaround for pytorch issue with serializing dtype by @ktschuett in #442
• Make compatible with older PyTorch versions by @ktschuett in #443
• maxstep can be used to renormalize the integration steps by @jnsLs in #441
• Add config docs by @ktschuett in #444
• NeighborListWrapper is removed by @jnsLs in #446
• Add SO3 ops and layers by @ktschuett in #445
• Moved and adapted CLI docs for MD. by @mgastegger in #447
• SO3 representation by @ktschuett in #448
• Fix ScaleProperty transform by @mgastegger in #450
• Add mixing layers to SO3net by @ktschuett in #451
• Fix load_properties by @NiklasGebauer in #454
• Fix metrics computation by @ktschuett in #455
• Separate metric objects for train/val/test by @ktschuett in #456
• Updates for response experiments / NMR utilities by @mgastegger in #457
• Cached NBL takes nbh_transforms by @jnsLs in #458
• Fix configs and docstrings by @ktschuett in #459
• Fix EMA, upgrade to PL 1.8 by @ktschuett in #460
• Refactored MD17 DataModule and added MD22 datset by @mgastegger in #461
• Readd residual by @ktschuett in #463
• Make return of vector representation in SO3net optional by @mgastegger in #465
• Make transforms independently usable from datamodule by @ktschuett in #464
• fixes initialization bug in so3net by @Stefaanhess in #467
• added ASEBatchwiseLBFGS, updated tutorial by @jnsLs in #466
• updated tutorial 5 by @Stefaanhess in #468
• Fixed typos in tutorial 5 by @mgastegger in #469
• Added reference to schnetpack-gschnet to docs... by @NiklasGebauer in #470
• cleaned up batchwise optimizer by @jnsLs in #471
• updated requirements.txt by @Stefaanhess in #475
• added trained model to tutorial by @Stefaanhess in #474
• Updated docs and fixed problems by @mgastegger in #477
• Switched to matscipy pypi package by @mgastegger in #478

New Contributors

• @Divide-By-0 made their first contribution in #311
• @jnsLs made their first contribution in #322
• @chgaul made their first contribution in #357
• @NiklasGebauer made their first contribution in #369
• @dependabot made their first contribution in #437

Full Changelog: v1.0.0...v2.0.0