v0.5.2

[0.5.2] - 2022-02-04

Added

• Model builders may now process only the configuration
• Allow irreps to optionally be specified through the simplified keys l_max, parity, and num_features
• wandb.watch via wandb_watch option
• Allow polynomial cutoff p values besides 6.0
• nequip-evaluate now sets a default r_max taken from the model for the dataset config
• Support multiple rescale layers in trainer
• AtomicData.to_ase supports arbitrary fields
• nequip-evaluate can now output arbitrary fields to an XYZ file
• nequip-evaluate reports which frame in the original dataset was used as input for each output frame

Changed

• minimal.yaml, minimal_eng.yaml, and example.yaml now use the simplified irreps options l_max, parity, and num_features
• Default value for resnet is now False

Fixed

• Handle one of per_species_shifts/scales being null when the other is a dataset statistc
• include_frames now works with ASE datasets
• no training data labels in input_data
• Average number of neighbors no longer crashes sometimes when not all nodes have neighbors (small cutoffs)
• Handle field registrations correctly in nequip-evaluate

Removed

• compile_model