API reference¶
This reference is generated from the docstrings in the icu_benchmarks package, which use the Google style.
| Module | Contents |
|---|---|
| Entry point and run utilities | The icu-benchmarks CLI, argument parsing, run directories, result aggregation, repeated cross-validation |
| Data | Splitting and preprocessing (preprocess_data), preprocessors, datasets, target scaling, pooling |
| Training and model wrappers | train_common, load_model, the ML/DL/imputation wrappers, metrics |
| Deep learning models | RNN, LSTM, GRU, GRU-D, TCN, Transformer, Bi-Axial Transformer, Mamba |
| Classical machine learning models | Logistic regression, elastic net, random forest, SVM, LightGBM, XGBoost, CatBoost, imbalanced-learn |
| Imputation models | Baselines, MLP, RNN, neural processes, CSDI, diffusion and SSSD models, amputation |
| Hyperparameter tuning | The Optuna tuner and its gin helpers |
Most classes and functions are gin configurables: any of their arguments can be set in a .gin file or with -hp on the
command line, as described in Configuration.