NEWS
IBLM 2.0.1
Major changes
- train_iblm_xgb() methodology changed. The booster model component is now trained differently, using glm predictions as an offset for xgb.train() instead of the ratios method employed in versions ≤ 1.0.3.
- Structure of data object freMTPLmini changed. Output of load_freMTPL2freq() also changed accordingly.
New features
- The IBLM model may now be trained using a weighting variable and/or an offset variable. More information is available at https://ifoa-adswp.github.io/IBLM/articles/IBLM.html#offsetting-and-weighting.
IBLM 1.0.2 (2025-12-16)
Major changes
iblm class model objects now require factor variables for categorical features. Users should explicitly convert character columns to factors before training.
New features
- Added "IBLM" vignette providing a comprehensive walkthrough of package functionality.
Bug fixes
- Fixed bug where all variable types were converted to numeric before training and predicting of booster model, which could lead to improper handling of categorical features.
Internal changes
- Updated to maintain compatibility with xgboost v3.1.2.1. Package now requires xgboost >= 3.1.2.1.
IBLM 1.0.0