XAI-metrics

A package for analysis and evaluating metrics for machine learning models explainability.

Installation

Install from PyPI:

pip install xai-metrics

Usage

Examples of usage:

• Perturbation based on permutation importances

from xai_metrics import examine_interpretation

X_train.columns = ['0','1','2','3']
X_test.columns = ['0','1','2','3']
xgb_model = xgb.XGBClassifier()
xgb_model.fit(X_train, y_train)
perm = PermutationImportance(xgb_model, random_state=1).fit(X_test, y_test)
perm_importances = perm.feature_importances_

examine_interpretation(xgb_model, X_test, y_test, perm_importances, epsilon=4, resolution=50, proportionality_mode=0)

• Perturbation based on local importances

from xai_metrics import examine_local_fidelity

examine_local_fidelity(xgb_model, X_test, y_test, epsilon=3)

• Gradual elimination

from xai_metrics import gradual_elimination

gradual_elimination(f_forest, f_X_test, f_y_test, f_shap)

See here for notebooks with full examples of usage.