xgboost in python and pyspark

xgboost in python and pyspark (using py4j to call jvm-packages)
xgboost4j version: 0.82

TODO: xgboost4j is not the latest version since 0.90 only supports python3 and spark 2.4

how to set environment (without docker)

1. download xgboost4j-0.82 jar files from xgboost-jars
2. copy to pyspark_xgb/jars
3. rename to xgboost4j-0.82.jar and xgboost4j-spark-0.82.jar respectively
4. set your SPARK_HOME and JAVA_HOME in pyspark/start.sh
5. [opt] change spark-submit parameters if needed

run xgboost

python version 2.7

• binary logistic

python python_xgb/train_binary.py

• multi classification

python python_xgb/train_multi.py

run xgboost4j (py4j to call function in xgboost jvm-packages)

spark version 2.3.*

• binary logistic

pyspark_xgb/start.sh train_binary.py

• multi classification

pyspark_xgb/start.sh train_multi.py

Appendix

run the program within docker

how to set environment (docker)

build images from docker file (~3GB)

it takes some time to build the images ...

cd docker
docker build -t xgb:latest . --no-cache

start docker container using images, go to project directory

docker run -i -t xgb:latest /bin/bash
cd xgboost-python-pyspark