Daniel Wlazło
Open-source projects, mostly around explainable and trustworthy machine learning. The full site — academic work, teaching, and conference talks — lives at datadeer.pl.
Projects
treecf
Constrained, threshold-aware counterfactual explanations for tree ensembles (XGBoost, LightGBM, CatBoost, scikit-learn), running on a bundled Rust engine.
probcal
Universal post-hoc probability calibration for binary classifiers: methods, metrics, diagnostics, and auditable offsetting — numpy-only. Built for credit-risk PD models, fully general.
triadxai
TRIAD: three-channel attribution decomposition (Information / Density / Missingness) for credit-risk models — exact for additive models, SHAP-based for black-box GBMs.
swift
SWIFT (SHAP-Weighted Impact Feature Testing): model-aware distribution monitoring for ML systems. Uses model decision boundaries for bucketing, SHAP values for normalization, and Wasserstein distance for comparison.
concept-graph-xai
Concept-graph aware visualisation of model feature usage and importance, with concept-level ablation metrics.
flaggam
Interpretable generalized additive models built from univariate feature flags.