- add citation information and zenodo badge
- add code to estimate the parameters for the platt-probabilty estimate
- allow specification of minimum occurrences of sub-substructures in the circular fingerprints as integer (minimum number)
- add support for sparse fingerprint string outputs for the circular fingerprinter
- Joblib is used to compute the fingerprints of multiple molecules in parallel
- Joblib uses the 'multiprocessing' backend to make it work
- Small performance improvements in the feature transformation function
- Sometimes SMILES strings cannot be parsed and converted to rdkit mol-objects because of "explicit-valence-errors"
- A new sanitization is added, trying to recover from the error and produce a valid mol-object.
- make featurizer compatible with GridSearchCV (only single core, i.e.
n_jobs=1
)
- improved convergence of the RankSVM by using line-search and early stopping criteria