For this release, we refactored configurations and script names such that the training and generation can be invoked and configured more conveniently. The README has been updated to reflect these changes.
We provide a new script called check_validity.py that allows to assess the validty, uniqueness, and novelty of generated molecules in a standardized way. It builds on the implementation of xyz2mol available in RDKit. See the README for instructions.
The model checkpoint now stores additional information on training settings, e.g. the distance unit used in the reference structures.
Furthermore, additional information about the settings chosen for molecule generation are stored in the data bases of generated molecules.
The changes might lead to errors when loading models trained with previous releases, so use previous versions to generate molecules with older models.