Skip to content

Latest commit

 

History

History
121 lines (98 loc) · 5.46 KB

README.md

File metadata and controls

121 lines (98 loc) · 5.46 KB
Raw

Spark Rapids ML (Scala)

PCA

Comparing to the original PCA training API:

val pca = new org.apache.spark.ml.feature.PCA()
  .setInputCol("feature_vector_type")
  .setOutputCol("feature_value_3d")
  .setK(3)
  .fit(vectorDf)

We used a customized class and user will need to do no code change to enjoy the GPU acceleration:

val pca = new com.nvidia.spark.ml.feature.PCA()
  .setInputCol("feature_array_type") // accept ArrayType column, no need to convert it to Vector type
  .setOutputCol("feature_value_3d")
  .setK(3)
  .fit(vectorDf)
...

Note: The setInputCol is targeting the input column of Vector type for training process in CPU version. But in GPU version, user doesn't need to do the extra preprocess step to convert column of ArrayType to Vector type, the setInputCol will accept the raw ArrayType column.

Build

Build in Docker:

We provide a Dockerfile to build the project in a container. See docker for more instructions.

Prerequisites:

  1. essential build tools:
  2. CUDA Toolkit(>=11.5)
  3. conda: use miniconda to maintain header files and cmake dependecies
  4. cuDF:
    • install cuDF shared library via conda: 
      conda install -c rapidsai -c conda-forge cudf=22.04 python=3.8 -y
  5. RAFT(22.12):
    • raft provides only header files, so no build instructions for it. Note we fix the version to 22.12 to avoid potential API compatibility issues in the future. 
      $ git clone -b branch-22.12 https://github.com/rapidsai/raft.git
  6. export RAFT_PATH: 
    export RAFT_PATH=ABSOLUTE_PATH_TO_YOUR_RAFT_FOLDER

Note: For those using other types of GPUs which do not have CUDA forward compatibility (for example, GeForce), CUDA 11.5 or later is required.

Build target jar

Spark-rapids-ml uses spark-rapids plugin as a dependency. To build the SNAPSHOT jar, user needs to build and install the denpendency jar rapids-4-spark first because there's no snapshot jar for spark-rapids plugin in public maven repositories. See build instructions to get the dependency jar installed.

User can also modify the pom file to use the release version spark-rapids plugin as the dependency. In this case user doesn't need to manually build and install spark-rapids plugin jar by themselves.

Make sure the rapids-4-spark is installed in your local maven then user can build it directly in the project root path with:

cd jvm
mvn clean package

Then rapids-4-spark-ml_2.12-24.04.1-SNAPSHOT.jar will be generated under target folder.

Users can also use the release version spark-rapids plugin as the dependency if it's already been released in public maven repositories, see rapids-4-spark maven repository for release versions. In this case, users don't need to manually build and install spark-rapids plugin jar by themselves. Remember to replace the dependency in pom file.

Note: This module contains both native and Java/Scala code. The native library build instructions has been added to the pom.xml file so that maven build command will help build native library all the way. Make sure the prerequisites are all met, or the build will fail with error messages accordingly such as "cmake not found" or "ninja not found" etc.

How to use

After the building processes, spark-rapids plugin jar will be installed to your local maven repository, usually in your ~/.m2/repository.

Add the artifact jar to the Spark, for example:

ML_JAR="target/rapids-4-spark-ml_2.12-24.04.1-SNAPSHOT.jar"
PLUGIN_JAR="~/.m2/repository/com/nvidia/rapids-4-spark_2.12/24.04.1/rapids-4-spark_2.12-24.04.1.jar"

$SPARK_HOME/bin/spark-shell --master $SPARK_MASTER \
 --driver-memory 20G \
 --executor-memory 30G \
 --conf spark.driver.maxResultSize=8G \
 --jars ${ML_JAR},${PLUGIN_JAR} \
 --conf spark.plugins=com.nvidia.spark.SQLPlugin \
 --conf spark.rapids.sql.enabled=true \
 --conf spark.task.resource.gpu.amount=0.08 \
 --conf spark.executor.resource.gpu.amount=1 \
 --conf spark.executor.resource.gpu.discoveryScript=./getGpusResources.sh \
 --files ${SPARK_HOME}/examples/src/main/scripts/getGpusResources.sh

PCA examples

Please refer to PCA examples for more details about example code. We provide both Notebook and jar versions there. Instructions to run these examples are described in the README.