dmlc/xgboost

Implements parallel tree boosting with built-in support for categorical features, missing value handling, and monotonic constraints without preprocessing. Uses a novel column-block structure for cache-aware tree construction and supports GPU acceleration via CUDA for faster training on large datasets. Integrates with ML platforms including scikit-learn, MLflow, and Optuna for hyperparameter optimization, with native support for feature importance analysis and SHAP explainability.

28,121 stars and 41,912,233 monthly downloads. Used by 120 other packages. Actively maintained with 45 commits in the last 30 days. Available on PyPI.