feature_importance

CatBoost feature-importance топ-N в формате Шелестова.

Усредняем PredictionValuesChange (native) по всем 5 сидам ансамбля, добавляем русское описание из feature_labels_ml.json. Пишем statistics/tables/feature_importance.csv:

rank,feature,importance,label

Запуск:

python ml/src/feature_importance.py \
    --artifacts ml/artifacts \
    --out statistics/tables/feature_importance.csv \
    --top 20

Functions

main()

feature_importance.main() → None