Releases · CDDLeiden/DrugEx

From v3.1.0 to v3.2.0

Fixes

fixes to SmilesExplorerNoFrag (wrong best state was saved and a TypeError while logging was eliminated, see !40)
optimized how memory is used a little bit (!50)
fix #55

generated SMILES are now not reported in the logger of SmilesExplorerNoFrag (see !40), but should still be available to the supplied training monitor
Training QSAR models is restructured (see !41), only CLI still environ.py, actually functionality moved to environment.
As well as unittests added for this part of the code.
Early stopping patience is now softcoded for all models and for reinforcement learning criteria was changed from ratio of desired molecules to the mean average score (see !46)

add option to remove molecules with tokens not occuring in voc (in dataset.py), see !39.
add grid search for DNN QSAR model (see !41)
add bayes optimization for DNN QSAR model (see !42)
add option to use different environment algorithms during RL
add option to use selectivity window predictor for RL
add option to specify chunk_size in the dataset.py script to control how data is supplied to parallel processes (bigger chunk size -> more memory used, but more efficient use of multiple CPUs, see !50)