How to use XGBoost from within Mathematica?

Question

Is there an easy and efficient way to use XGBoost from within Mathematica without naively Import/Exporting the data? Does anyone know of a third party package or any example applications of integrating with XGBoost?

Details:

• Nvidia's blog about their support for using XGBoost with their ML primitives and containers https://devblogs.nvidia.com/gpu-accelerated-analytics-rapids/
• Lots of data science workloads that take huge amounts of time on Python (let alone MMA) are much faster with https://rapids.ai/

Answer 1

Needs["JLink`"]
SetOptions[InstallJava, JVMArguments -> "-Xmx3g"];

Needs["RLink`"]
InstallR["RHomeLocation" -> "C:\\R\\R-3.4.4"]
REvaluate["library(\"xgboost\")"]

{"xgboost", "stats", "graphics", "grDevices", "utils", "datasets", "methods", "base"}

SeedRandom[0];
RSet["x", RandomReal[{-1, 1}, {1000, 50}]];
RSet["y", N@RandomInteger[{0, 1}, 1000]];

REvaluate["set.seed(0)"];
REvaluate["bst<-xgboost(
       data=x,
       label=y,
       booster=\"gbtree\",
       nthread=4,
       objective=\"binary:logistic\",
       eval_metric=\"auc\",
       nround=559,
       max_depth=5,
       eta=0.02,
       subsample=0.7,
       colsample=0.7
       )"];

RSet["z", RandomReal[{-1, 1}, {5, 50}]];
REvaluate["predict(bst,z,ntreelimit=559)"]

{0.608143, 0.677535, 0.4781, 0.60045, 0.502672}