[jira] [Commented] (SPARK-15720) MLLIB Word2Vec loading large number of vectors in the model results in java.lang.NegativeArraySizeException

["yuhao yang (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-15720?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15311647#comment-15311647 ] 

yuhao yang commented on SPARK-15720:
------------------------------------

This can only happen when creating a Word2VecModel from pre-trained vectors, as currently Word2Vec in MLlib cannot support that scope. Current upper limit is vocab * vectorSize < Max Array Size (approximately (Int.Max -  8) depending on different platforms).

This will be a fundamental change to Word2Vec if we want to extend the scope. [~rohangpatil] What's the target scope of vocabSize and vectorLength for your application. I'm not sure if larger scope of word2vec is a popular requirement, appreciate if you can provide some supporting examples. 

> MLLIB Word2Vec loading large number of vectors in the model results in java.lang.NegativeArraySizeException
> -----------------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-15720
>                 URL: https://issues.apache.org/jira/browse/SPARK-15720
>             Project: Spark
>          Issue Type: Bug
>          Components: MLlib
>    Affects Versions: 1.6.1
>         Environment: Linux
>            Reporter: Rohan G Patil
>
> While loading a large number of pre-trained vectors into Spark MLLIB's Word2Vec model, will result in java.lang.NegativeArraySizeException.
> Code - https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/feature/Word2Vec.scala#L597
> Test with number of vectors greater than 16777215 with size of each vector 128 or more.
> there is Integer Overflow happening here. Should be an easy fix.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

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