XGBoost is an open-source library that offers gradient boosting which is an ensemble method. The algorithm runs on a notebook or scales to support multiple cores, GPUs, threads, and clusters of computing. Supports scikit-learn but has its own interface. 

Project Background

• Library: XGBoost
• Author: Tianqi Chen and DMLC group
• Initial Release: March 2014
• Type: Gradient boosting library
• License: Apache 2.0
• Language: Developed in C++
• GitHub: dmlc/xgboost
• Runs On: Single machine and supports Hadoop, Spark, and Flink
• Twitter: XGBoost

Features

• Boosting is an ensemble method
• The algorithm wins lots of Kaggle competitions
•  Supports regularized boosting which is normalized
• Handles missing values automatically
• Takes advantage of multiple cores, threads, and clusters of compute
•  Incremental training: stop and save training, then come back to it later
• Tree pruning: deeper and more optimal trees
• 3 types of boosting supported: gradient, stochastic gradient, and regularized boosting