Feature importances

The importance of each input feature is its share of the model's variance: each table's share is split equally among the features it involves. The importances are read from the model, need no data, and sum to 1. See Feature importance for the calculation principles.

Python package

Attribute

• feature_importances_ (TBoostRegressor)
• feature_importances_ (TBoostClassifier)

Usage examples

for name, importance in zip(model.feature_names_in_, model.feature_importances_):
    print(f"{name}: {importance:.3f}")

The share of each table is in the sobol field of the exported tables.