Releases: RekerLab/MolALKit
v0.9.1
v0.9.0
Update to be compatible with mgktools==2.0.0.
Bugfix for chemprop models and GPR-MGK-Morgan when dealing with multi-molecular datasets.
v0.8.0
Add surrogate ML models: adaboost and xgboost.
Add features: sequential-pool active learning; partial-query active learning.
v0.7.0
Support logger output of chemprop.
Support continuous_fit of chemprop during active learning.
Support read pretrained model for chemprop.
v0.6.3
Simplify the code by using the args of chemprop for DMPNN.
Set default training epoch to 30 for neural networks.
Bug fix:
Set the acquisition value of passive learning to [] instead of None.
Missing trajectory frames for checkpoint file.
Add argument:
--write_traj_stride, set the frequency to output the trajectory file.
v0.6.2
Bugfix:
Continue active learning from checkpoint file.
Do not write repeated results when continue active learning from checkpoint file.
v0.6.1
Bugfix:
The parameter C of support vector machine is not correctly passed.
The model configs for two molecules data sets are updated.
v0.6.0
Add 3 metrics for regression tasks: Spearman's r, Pearson's r, and Kendall's tau.
v0.5.2
Bugfix:
- Evaluate the model performance at the last iteration.
- Unify the dataset structure of mgktools, molalkit, and deepchem. The dataset.y is always 2-dimensional numpy array.
- Add more unit tests.
v0.5.1
bugfix: when init_size is larger than the entire data, turn to cross validation.