RapidMiner XGBoost Extension

This extension embeds the XGBoost eXtreme Gradient Boosting library for use in RapidMiner. It implements a single operator named XGBoost compatible with RapidMiner's builtin learners.

Features

Automatic data conversion

• Automatic data conversion between RapidMiner tables and XGBoost's internal format. Supports all data types except datetime values.

Encoding of categorical data

• Automatic encoding of categorical features using binary columns or a modified one-hot encoding. The user can of course always implement their own encoding using RapidMiner operators.

Objective selection

• Automatic selection of the learning objective based on the input data. Supports regression, binary classification, and multi-categorical classification problems. The objective can also be specified explicitly by the user.

Limitations

No GPU support

• The distribution of XGBoost used by this extension lacks GPU support. Thus, the extension only provides the CPU backends of XGBoost.

Limited concurrency

• The extension limits the concurrent use of XGBoost learners due to observed instabilities of some XGBoost APIs. Take note that a single XGBoost learner will still scale out to multiple CPU cores. However, running multiple XGBoost learners in parallel on a many-core system might underutilize the CPU.