Ensemble-Pytorch

• Official support for Python 3.9 / 3.10 | @Malephilosopher
• Minor bug fix | @xuyxu

• Add NeuralForestClassifier and NeuralForestRegressor | @xuyxu
• Fix the sampling issue in BaggingClassifierandBaggingRegressor` | @SunHaozhe and @FedericoV
• Relax check on input dataloader | @xuyxu

• Support arbitary arbitrary training criteria for all models except gradient boosting
• Minor bug fix

• Fix missing functionality of param save_model for soft gradient boosting | @xuyxu
• Add SoftGradientBoostingClassifier and SoftGradientBoostingRegressor | @xuyxu

• Supoort using dataloader with multiple input
• Large number of doc improvements

• Add SoftGradientBoostingClassifier and SoftGradientBoostingRegressor | @xuyxu
• Lots of doc improvements | @xuyxu

• Fix missing functionality of use_reduction_sum for fit of Gradient Boosting | @xuyxu
• Relax tensorboard as a soft dependency | @xuyxu
• Simplify the training workflow of FastGeometricClassifier and FastGeometricRegressor | @xuyxu
• Support TensorBoard logging in set_logger | @zzzzwj
• Add use_reduction_sum parameter for fit of Gradient Boosting | @xuyxu
• Improve the functionality of evaluate and predict | @xuyxu

Added

• Add FastGeometricClassifier and FastGeometricRegressor | @xuyxu
• Add flexible instantiation of optimizers and schedulers | @cspsampedro
• Add support on accepting instantiated base estimators as valid input | @xuyxu

Fixed

• Fix typos in Gradient Boosting | @nolaurence

Added

• Add methods on model deserialization load for all ensembles | @mttgdd

Fixed

• Fix missing base estimators when calling load for all ensembles | @xuyxu

Added

• Add set_scheduler for all ensembles | @xuyxu
• Add AdversarialTrainingClassifier and AdversarialTrainingRegressor | @xuyxu
• Add SnapshotEnsembleClassifier and SnapshotEnsembleRegressor | @xuyxu
• Add model validation and serialization | @ozanpkr and @xuyxu
• Add CI and maintenance tools | @xuyxu
• Add the code coverage on codecov | @xuyxu
• Add the version numbers to requirements.txt | @zackhardtoname and @xuyxu
• Improve the logging module usinglogging | @zzzzwj

Changed

• Remove the input argument output_dim from all methods | @xuyxu
• Refactor the setup on optimizer into set_optimizer | @xuyxu
• Refactor the codes on operating tensors into an independent module | @zzzzwj

Fixed

• Fix the bug in logging module when using multi-processing | @zzzzwj
• Fix the binding problem on scheduler and optimizer when using parallelization | @Alex-Medium and @xuyxi