The original dataset consists of several clusters generated by different linear laws.
- For each point in the test set, determine the vector of feature importances
- Cluster the resulting importance vectors
- Determine the law by which each of the clusters are generated
ML Models:
- XGBoost (Extreme gradient boost) with GridSearchCV hyperparameter tuning
Files:
notebook: xai_clusters_gen_xgboost.ipynb
model: xgb_clusters_gen_v1.pkl - LGBM (Light gradient boost machines) with GridSearchCV hyperparameter tuning
Files:
notebook: xai_clusters_gen_lgbm.ipynb
model: lgbm_clusters_gen_v1.pkl - NAF (Neural Attention Forest) without hyperparameter tuning
Files:
notebook: xai_clusters_gen_naf.ipynb (just global explanations)
X-AI Algorithms:
- Tree-SHAP (Shapley additive explanation) for global and local explanations
X-AI Visualizations:
- For global explanations
- Feature importance bar plots
- Bar plots with hierarchical clustering among features
- Summary plots
- Heat maps
- Partial Dependency Plots (PDP)
- For local explanations
- Force plots
- Decision plots
- Waterfall plots