Changes

v0.6.2 ~ v0.6.3

Bug fix

  • some minor fixes.

v0.6.1

Breaking change

  • GaussianNLLLoss has been removed from xenonpy.model.training.loss. ( #249 )

v0.6.0

Bug fix

  • Fix a bug in the Scaler class. ( #243 )

New features

  • Add average option to the classification_metrics function. Now users can decide how to calculate scores for multilabel tasks. ( #240 )

  • Add only_best_states option to the Persist class. If True, Persist will only save the best state to reduce the storage space. ( #233 )

  • Add warming_up option to the Validator class. ( #238 )

v0.5.2

Bug fix

  • some minor fixes.

v0.5.1

Bug fix

  • update the pip install dependencies in requirements.txt. ( #226 )

Enhance

  • Add rdkit v2020.09 support for fingerprint descriptors. ( #228 )

  • Add return probability support for model.extension.TensorConverter. ( #229 )

v0.5.0

Breaking change

  • Replace xenonpy.datatools.BoxCox with xenonpy.datatools.PowerTransformer. ( #222 )

New features

  • Add xenonpy.datatools.PowerTransformer to provide yeo-johnson and box-cox transformation through the sklearn.preprocessing.PowerTransformer. ( #222 )

  • Add new contribution ISMD by Qi, a new class of BaseProposal that allows generation of molecules based on virtual reaction of reactants in a predefined reactant pool. ( #208 )

  • Add classifier training support for the xenonpy.model.Trainer. ( #184 )

  • Add IQSPR4DF to support pandas.DataFrame input to iQSPR.

  • Add LayeredFP, PatternFP, and MHFP (new rdkit fingerprints).

Enhance

  • BaseFeaturizer.transform now supports pandas.DataFrame as input where relevant columns for descriptor calculation can be specified through target_col.

  • BaseDescriptor.transform and BaseLogLikelihoodSet.log_likelihood now automatically check if any group names occur in the input pandas.DataFrame column names. If not, the entire pandas.DataFrame will be passed to the corresponding BaseFeaturizer and BaseLogLikelihood, respectively.

  • Allow using custom elemental information matrix in xenonpy.descriptor.Compositions descriptor. ( #221 )

  • Use joblib.parallel as default parallel backend. ( #191, #220 )

  • Splliter.split method now support python list as input. ( #194 )

  • Allow user specific index for DescriptorHeatmap. ( #44 )

  • Allow control of number of layers to be extracted in FrozenFeaturizer. ( #174 )

  • bit_per_entry option is added to RDKitFP and AtomPairFP to allow control of number of bits to represent one fingerprint entry.

  • counting option is added to RDKitFP, AtomPairFP, TopologicalTorsionFP, FCFP and ECFP to support returning counts of each fingerprint entry.

  • Column names of DescriptorFeature is updated to be consistent with the rdkit naming.

Infrastructure improve

  • Move CI to github action. ( #195 )

  • Move to readthedocs version 2. ( #206 )

v0.4.2

Bug fix

  • fix set_param method dose not set the children in BaseDescriptor and NGram. ( #163, #159 )

Enhance

  • Setting optimizer, loss_func, and etc. can be done in Trainer.load. ( #158 )

  • Improve docs. ( #155 )

v0.4.0

Breaking change

  • Remove xenonpy.datatools.MDL.

  • Remove xenonpy.model.nn modules. Part of them will be kept until v1.0.0 for compatible.

New features

  • Add xenonpy.mdl modules for XenonPy.MDL access.

  • Add xenonpy.model.training modules for model training.

v0.3.6

Breaking change

  • Renamed BayesianRidgeEstimator to GaussianLogLikelihood.

  • Removed the estimators property from BayesianRidgeEstimator.

  • Added predict method into GaussianLogLikelihood.

v0.3.5

Enhanced

  • Added version specifiers to the requirements.txt file.

v0.3.4

Bug fix

  • Fixed a critical error in BayesianRidgeEstimator when calculating the loglikelihood. ( #124 )

v0.3.3

Bug fix

  • fix mp_ids.txt not exist error when trying to build the sample data using preset.build.

v0.3.2

Enhanced

  • Updated sample codes.

  • Added progress bar for ngram training. ( #93 )

  • Added error handling to NGram when generating new SMILES. ( #97 )

CI

  • Removed python 3.5 support. ( #95 )

  • Added Appveyor CI for windows tests. ( #90 )

v0.3.1

Enhanced

  • Added tutorials for main modules. ( #79 )

v0.3.0

Breaking changes:

  • Removed Built-in data mp_inorganic, mp_structure, oqmd_inorganic and oqmd_structure. ( #12, #20 )

  • Renamed LocalStorage to Storage.

Enhanced

  • Added error handling for NGram training. ( #75, #86 )

  • Added error handling for IQSPR. ( #69 )

  • Added error handling for BaseDescriptor and BaseFeaturizer. ( #73 )

  • Added featurizer selection function. ( #47 )

New Features

  • Added sample data building function for preset. ( #81, #84 )

v0.2.0

Descriptor Generator:

  • Added xenonpy.descriptor.Fingerprint descriptor generator. ( #21 )

  • Added xenonpy.descriptor.OrbitalFieldMatrix descriptor generator. ( #22 )

API Changes:

  • Allowed BaseDescriptor class to use anonymous/renamed input. ( #10 )