Repository navigation
Documentation release.
Changed
- README: added CatBoost selection guidance to the agent-ensemble section — tool-scout's practical recommendation from the Telco Churn validation run. Prefer CatBoost on categorical-heavy data (many high-cardinality categorical columns, few numeric features): native categorical handling and ordered target encoding that guards against leakage, often edging out LightGBM at the cost of slower training. Keep LightGBM for large mostly-numeric data or when training speed matters; dispatch tool-scout to verify the best fit for your specific data.
Full changelog: see CHANGELOG.md