[
https://issues.apache.org/jira/browse/SPARK-37913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17795697#comment-17795697
]
APeng Zhang commented on SPARK-37913:
-------------------------------------
I can reproduce this issue.
h2. Solution:
A simple approach to resolve this issue is to extract
_UnaryTransformerExample.MyTransformer_ from {_}UnaryTransformerExample{_},
transforming it into a *top-level class* instead of a *nested class.*
h2. Code to reproduce the issue:
Adding these 2 blocks in _org.apache.spark.examples.ml.UnaryTransformerExample_
{code:java}
val pipeline = new Pipeline().setStages(Array(myTransformer))
val pipelinePath = "~/tmp/ts/1";
pipeline.write.overwrite().save(pipelinePath)
val pipeline2 = Pipeline.load(pipelinePath){code}
{code:java}
val model = pipeline.fit(data)
val pipelineModelPath = "~/tmp/ts/2";
model.write.overwrite().save(pipelineModelPath)
val model2 = PipelineModel.load(pipelineModelPath){code}
there will be the exception:
{code:java}
23/12/12 19:20:40 ERROR Instrumentation: java.lang.NoSuchMethodException:
org.apache.spark.examples.ml.UnaryTransformerExample$MyTransformer.read() at
java.base/java.lang.Class.newNoSuchMethodException(Class.java:660) at
java.base/java.lang.Class.throwExceptionOrReturnNull(Class.java:1458) at
java.base/java.lang.Class.getMethodHelper(Class.java:1544) at
java.base/java.lang.Class.getMethod(Class.java:1450) at
org.apache.spark.ml.util.DefaultParamsReader$.loadParamsInstanceReader(ReadWrite.scala:631)
at
org.apache.spark.ml.Pipeline$SharedReadWrite$.$anonfun$load$4(Pipeline.scala:276)
at scala.collection.ArrayOps$.map$extension(ArrayOps.scala:934) at
org.apache.spark.ml.Pipeline$SharedReadWrite$.$anonfun$load$3(Pipeline.scala:274){code}
h2. Root cause:
_Pipeline.PipelineWriter_ and _PipelineModel.PipelineModelWriter_ will persist
the *class name* of each stage when saving _Pipeline_ or {_}PipelineModel{_}.
In this example, the class name is
{*}_org.apache.spark.examples.ml.UnaryTransformerExample$MyTransformer_{*}. The
class name will be used by _PipelineReader_ and _PipelineModelReader_ to
construct the Class object which represents the class, then invoke the
_*read()*_ method.
The problem here is for the *nested class*
{_}UnaryTransformerExample.MyTransformer{_}, the Scala compiler will generate 2
different Java classes, the _*read()*_ method which is part of the Scala
companion object will be a member of Java class
{_}*org.apache.spark.examples.ml.UnaryTransformerExample$MyTransformer{color:#FF0000}${color}*{_},
it is not a member of
{_}*org.apache.spark.examples.ml.UnaryTransformerExample$MyTransformer*{_}. So
there is the exception "{_}java.lang.NoSuchMethodException:
org.apache.spark.examples.ml.UnaryTransformerExample$MyTransformer.read(){_}"
when loading the object with _PipelineReader_ and {_}PipelineModelReader{_}.
Note there is no such issue for a {*}non-nested class{*}, since the Scala
compiler will generate only one Java class.
h2. Java classes generated by Scala compiler:
{code:java}
public final class UnaryTransformerExample {
...
public static class MyTransformer extends UnaryTransformer<Object, Object,
MyTransformer> implements DefaultParamsWritable {
...
public MLWriter write() {
return DefaultParamsWritable.write$(this);
}
public void save(String path) throws IOException {
MLWritable.save$((MLWritable)this, path);
}
...
}
public static class MyTransformer$ implements
DefaultParamsReadable<MyTransformer>, Serializable {
public MLReader<UnaryTransformerExample.MyTransformer> read() {
return DefaultParamsReadable.read$(this);
}
public Object load(String path) {
return MLReadable.load$((MLReadable)this, path);
}
...
}
}
{code}
> Null Pointer Exception when Loading ML Pipeline Model with Custom Transformer
> -----------------------------------------------------------------------------
>
> Key: SPARK-37913
> URL: https://issues.apache.org/jira/browse/SPARK-37913
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 3.1.2
> Environment: Spark 3.1.2, Scala 2.12, Java 11
> Reporter: Alana Young
> Priority: Critical
> Labels: MLPipelineModels, MLPipelines
>
> I am trying to create and persist a ML pipeline model using a custom Spark
> transformer that I created based on the [Unary Transformer
> example|https://github.com/apache/spark/blob/v3.1.2/examples/src/main/scala/org/apache/spark/examples/ml/UnaryTransformerExample.scala]
> provided by Spark. I am able to save and load the transformer. When I
> include the custom transformer as a stage in a pipeline model, I can save the
> model, but am unable to load it. Here is the stack trace of the exception:
>
> {code:java}
> 01-14-2022 03:49:52 PM ERROR Instrumentation: java.lang.NullPointerException
> at java.base/java.lang.reflect.Method.invoke(Method.java:559) at
> org.apache.spark.ml.util.DefaultParamsReader$.loadParamsInstanceReader(ReadWrite.scala:631)
> at
> org.apache.spark.ml.Pipeline$SharedReadWrite$.$anonfun$load$4(Pipeline.scala:276)
> at
> scala.collection.TraversableLike.$anonfun$map$1(TraversableLike.scala:238) at
> scala.collection.IndexedSeqOptimized.foreach(IndexedSeqOptimized.scala:36) at
> scala.collection.IndexedSeqOptimized.foreach$(IndexedSeqOptimized.scala:33)
> at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:198) at
> scala.collection.TraversableLike.map(TraversableLike.scala:238) at
> scala.collection.TraversableLike.map$(TraversableLike.scala:231) at
> scala.collection.mutable.ArrayOps$ofRef.map(ArrayOps.scala:198) at
> org.apache.spark.ml.Pipeline$SharedReadWrite$.$anonfun$load$3(Pipeline.scala:274)
> at
> org.apache.spark.ml.util.Instrumentation$.$anonfun$instrumented$1(Instrumentation.scala:191)
> at scala.util.Try$.apply(Try.scala:213) at
> org.apache.spark.ml.util.Instrumentation$.instrumented(Instrumentation.scala:191)
> at org.apache.spark.ml.Pipeline$SharedReadWrite$.load(Pipeline.scala:268) at
> org.apache.spark.ml.PipelineModel$PipelineModelReader.$anonfun$load$7(Pipeline.scala:356)
> at org.apache.spark.ml.MLEvents.withLoadInstanceEvent(events.scala:160) at
> org.apache.spark.ml.MLEvents.withLoadInstanceEvent$(events.scala:155) at
> org.apache.spark.ml.util.Instrumentation.withLoadInstanceEvent(Instrumentation.scala:42)
> at
> org.apache.spark.ml.PipelineModel$PipelineModelReader.$anonfun$load$6(Pipeline.scala:355)
> at
> org.apache.spark.ml.util.Instrumentation$.$anonfun$instrumented$1(Instrumentation.scala:191)
> at scala.util.Try$.apply(Try.scala:213) at
> org.apache.spark.ml.util.Instrumentation$.instrumented(Instrumentation.scala:191)
> at
> org.apache.spark.ml.PipelineModel$PipelineModelReader.load(Pipeline.scala:355)
> at
> org.apache.spark.ml.PipelineModel$PipelineModelReader.load(Pipeline.scala:349)
> at org.apache.spark.ml.util.MLReadable.load(ReadWrite.scala:355) at
> org.apache.spark.ml.util.MLReadable.load$(ReadWrite.scala:355) at
> org.apache.spark.ml.PipelineModel$.load(Pipeline.scala:337) at
> com.dtech.scala.pipeline.PipelineProcess.process(PipelineProcess.scala:122)
> at com.dtech.scala.pipeline.PipelineProcess$.main(PipelineProcess.scala:448)
> at com.dtech.scala.pipeline.PipelineProcess.main(PipelineProcess.scala) at
> java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native
> Method) at
> java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
> at
> java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
> at java.base/java.lang.reflect.Method.invoke(Method.java:566) at
> org.apache.spark.deploy.worker.DriverWrapper$.main(DriverWrapper.scala:65) at
> org.apache.spark.deploy.worker.DriverWrapper.main(DriverWrapper.scala){code}
>
> *Source Code*
> [Unary
> Transformer|https://gist.github.com/ally1221/ff10ec50f7ef98fb6cd365172195bfd5]
> [Persist Unary Transformer & Pipeline
> Model|https://gist.github.com/ally1221/42473cdc818a8cf795ac78d65d48ee14]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
