- Renamed project to better reflect the expanded functionality of the toolbox! 🔥
- New Pruner class: simpler, more homogenized implementation of iterative subset pruning, sharing the same basic architecture across different similarity evaluation functions, with cached calls to the eval similarity function.
- Added new debug dedicated log, that collects all debug messages if the DEBUG keyword is specified.
- Fixed bug with some parts of torsion_module for segmented graphs, mainly for rotationally corrected RMSD calculation.
- General code debug and PEP 8 cleanup (thanks Ruff).
- Informative suggestions: keywords that are not understood can generate suggestion of known keywords to use instead.
- Removed rmsd library dependency for local numba numpy implementation.
- Increased similarity_refining thresholds for the maximum number of structures to be compared: 1E5 for RMSD and MOI, 1E4 for rotationally corrected RMSD. Added printouts in log if we skip a pruning stage because the ensemble is too large.
- Fixed a bug in _get_rotation_mask that would rotate the wrong part of the molecule (spotted because prune_by_rmsd_rot_corr would not reject some molecules that it should have)
- Added a debug messages printout function for the Pruner class, which by default is set to embedder.debuglog
- Added support for AIMNET2 Pytorch models, including GPU support, with the package aimnet2-firecode. If a solvent is specified, the ALPB ΔGsolv (XTB) is added to the vacuum SP EE.
- Removed support for the THREADS keyword and command line setting, as the number of workers in every multithreaded parallelism is now automated.
TSCoDe Changelog (old)