New Features

• Polars backend for preprocessing (significantly faster preprocessing)
• Modality selection (for doing experiments that need to show modality contributions)
• Optuna optimization with visualization and checkpointing
• Options to use SHAP or other explainers (improved support pending)
• Introduced configuration settings for multiple classifiers including Balanced Random Forest, CatBoost, XGBoost, and RUSBClassifier.
• Refactored RNN, GRU, TCN, and Transformer.
• Enhanced capabilities for binary classification, regression, and dataset imputation tasks.
• Added new configuration for benchmarking experiments.
• Enhanced error handling in cross-validation processes.

Chores

• Updated dependencies and package versions for better compatibility.

What's Changed

• Requirements and setup update by @rvandewater in #120
• reformat by @rvandewater in #121
• Pypi readme fixes by @rvandewater in #122
• Recipies update by @rvandewater in #124
• Dataset fix and small setup/readme adjustments by @rvandewater in #126
• Use correct (?) scaling in DefaultRegressionPreprocessor by @mlondschien in #127
• Fix commands flags by @HendrikSchmidt in #133
• logging fix by @rvandewater in #134
• Added metrics by @rvandewater in #135
• Fix regular ML model training by @rvandewater in #132
• comment in on_test_epoch_start by @rvandewater in #137
• By default, use quicker versions of nafill recently added to ReciPys by @prockenschaub in #139
• Pooled experiments by @rvandewater in #140
• Transfer learning by @rvandewater in #136
• Rename kf to kidney_function by @prockenschaub in #145
• Hyperparameter opt by @rvandewater in #147
• Bump pyarrow from 11.0.0 to 14.0.1 by @dependabot in #148
• Bump scikit-learn from 1.2.2 to 1.5.0 by @dependabot in #149
• Bump black from 24.2.0 to 24.3.0 by @dependabot in #150
• Bump tqdm from 4.64.1 to 4.66.3 by @dependabot in #151
• Polars by @rvandewater in #152
• Bump lightning from 2.3.0 to 2.3.3 by @dependabot in #156
• Version release based on practice Cassandra use by @rvandewater in #155

New Contributors

• @mlondschien made their first contribution in #127
• @dependabot made their first contribution in #148

Full Changelog: v0.3.0...v1.0.0

The public release contains a lot of fixes and refactoring.

This version of YAIB adds Pytorch Lightning training, Wandb logging, and a range of imputation methods.

What's Changed

• Imputation Models, Processing, and Refactorings
• Refactoring updates

Full Changelog: latest...v0.2.0

First public release of Yet Another ICU Benchmark (YAIB).