version 0.6.0

0.6.0 [21/06/2021]

New

• Add interface to scikit regressors (#85)
• Add interface to HDF5 to store the training results (#88)
• Add interface to GPyTorch (#90)

Changed

• Fix prediction functionality (#81)
• Return predicted and expected values scaled back (#90)

version 0.5.0

Changed

• Fix graph neural network implementation (#59)
• Rename the graph neural network to MPNN (message passing NN)

New

• Interface to se3-transformer (#57)
• Calculate a guess to the molecular coordinates using a force field optimization
• Add bond distance as a feature for the GNN using the optimized geometries (#59)
• Add early stopping functionality

version 0.4.0

0.4.0 [02/10/2020]

Changed

• Removed duplicate CAT functionality (#36)
• Moved the properties computation and filtering to its own repo(#44, #45)

Version 0.3.0

0.3.0 [21/08/2020]

New

• Introduce Pipeline to filter ligands (#13, #26)
• Use SCScore
• Use Horovod to distribute the training
• Add mypy test

Version 0.2.0

Swan is a Python package to create statistical models to predict molecular properties.

New

• Allow to train in GPU.
• Use Pytorch-geometric to create molecular graph convolutional networks.

Changed

• Replace deepchem with pytorch
• Use rdkit to compute the fingerprints.