Veritas is a versatile verification tool for tree ensembles. You can use Veritas to generate adversarial examples, check robustness, find dominant attributes or simply ask domain specific questions about your model.
Install from PyPI:
pip install dtai-veritas
From source:
# Clone this repository
git clone https://github.com/laudv/veritas.git
cd veritas
git submodule init && git submodule update
pip install .
Veritas should work on Linux (GCC), Mac (LLVM), and Windows (MSVC). If you encounter issues, feel free to contact me or open an issue.
To pull the latest updates from Github, simply git pull
the changes and reinstall using pip
: pip install --force-reinstall .
.
You can convert an existing ensemble using the veritas.get_addtree
function for XGBoost, LightGBM and scikit-learn.
Here's an example of a model trained by XGBoost that has been converted to Veritas' own tree ensemble representation.
!code PART get_addtree_example!
The output is an AddTree consisting of 3 trees, as was defined in the XGBClassifier.
!output PART get_addtree_example!
Versatile Verification of Tree Ensembles. Laurens Devos, Wannes Meert, and Jesse Davis. ICML 2021 http://proceedings.mlr.press/v139/devos21a.html
Robustness Verification of Multiclass Tree Ensembles. Laurens Devos, Lorenzo Cascioli, and Jesse Davis. AAAI 2024