[ 
https://issues.apache.org/jira/browse/SPARK-15526?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16071392#comment-16071392
 ] 

Villu Ruusmann commented on SPARK-15526:
----------------------------------------

This issue is responsible for >95% of support requests around my JPMML-SparkML 
and JPMML-SparkML-Package projects.

I've given up teaching people how to use Apache Maven Shade plugin to work 
around it. For Apache Spark 2.0 and 2.1 users the "official fix" is to delete 
the two offending JAR files `$SPARK_HOME/jars/pmml-model-1.2.15.jar` and 
`$SPARK_HOME/jars/pmml-schema-1.2.15.jar`. Sure, it will break built-in MLlib 
PMML export functionality, but anyone who is interested in using their 
JPMML-SparkML replacements, shouldn't mind it.

> Shade JPMML
> -----------
>
>                 Key: SPARK-15526
>                 URL: https://issues.apache.org/jira/browse/SPARK-15526
>             Project: Spark
>          Issue Type: Dependency upgrade
>          Components: ML, MLlib
>    Affects Versions: 2.0.0
>            Reporter: Villu Ruusmann
>            Priority: Minor
>   Original Estimate: 2h
>  Remaining Estimate: 2h
>
> The Spark-MLlib module depends on the JPMML-Model library 
> (org.jpmml:pmml-model:1.2.7) for its PMML export capabilities. The 
> JPMML-Model library is included in the Apache Spark assembly, which makes it 
> very difficult to build and deploy competing PMML exporters that may wish to 
> depend on different versions (typically much newer) of the same library.
> JPMML-Model library classes are not part of Apache Spark public APIs, so it 
> shouldn't be a problem if they are relocated by prepending a prefix 
> "org.spark_project" to their package names using Maven Shade Plugin. The 
> requested treatment is identical to how Google Guava and Jetty dependencies 
> are shaded in the final assembly.
> This issue is raised in relation to the JPMML-SparkML project 
> (https://github.com/jpmml/jpmml-sparkml), which provides PMML export 
> capabilities for Spark ML Pipelines. Currently, application developers who 
> wish to use it must tweak their application classpath, which assumes 
> familiarity with build internals.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org

Reply via email to