Maintained with ☕️ by
IcePanel logo

New version of SageMaker XGBoost algorithm available

Share

Services

Customers can now use a new version of the [SageMaker XGBoost](https://docs.aws.amazon.com/en%5Fpv/sagemaker/latest/dg/xgboost.html) algorithm that is based on version 0.90 of the open-sourced [XGBoost](https://github.com/dmlc/xgboost) framework. XGBoost is a highly efficient and flexible algorithm for problems in regression, classification, and ranking. In addition to the [new features brought by the v0.9 framework](https://github.com/dmlc/xgboost/releases/tag/v0.90), this new SageMaker release offers customers: * **Flexibility**: This new release of XGBoost algorithm can be used as a built-in algorithm or as a framework. Customers migrating from the previous built-in algorithms will only need to make a small change and specify the algorithm version when they call get\_image\_uri to make sure they get the latest version. Customers using it as a framework will have the flexibility to specify how they want to train by providing their own training scripts. * **Scalability**: this version has an improved distribution mechanism which resulted in significantly improved memory footprint, smoother experience training on larger clusters and on larger datasets. * **Extensibility**: customers can extend the [container image](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-algo-docker-registry-paths.html) by installing new packages, adding custom scripts, or extend the [container SDK](https://github.com/aws/sagemaker-xgboost-container) that SageMaker created to build algorithms. They can also train the algorithm on their local machine. To learn more, see documentation [here](https://docs.aws.amazon.com/en%5Fpv/sagemaker/latest/dg/xgboost.html).