[jira] [Commented] (SPARK-8708) MatrixFactorizationModel.predictAll() populates single partition only

["Antony Mayi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-8708?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14611304#comment-14611304 ] 

Antony Mayi commented on SPARK-8708:
------------------------------------

I suspect this issue with all data in single partition is also causing following of my problems:

At some point (seems to be random) my whole execution gets stuck after one of .predictAll() calls and there is a message in errorlog saying:

bq. 15/07/01 21:54:15 INFO shuffle.ShuffleMemoryManager: Thread 16468 waiting for at least 1/2N of shuffle memory pool to be free

I then have to kill the processing. Is there any simple workaround for this?

> MatrixFactorizationModel.predictAll() populates single partition only
> ---------------------------------------------------------------------
>
>                 Key: SPARK-8708
>                 URL: https://issues.apache.org/jira/browse/SPARK-8708
>             Project: Spark
>          Issue Type: Bug
>          Components: MLlib
>    Affects Versions: 1.3.0
>            Reporter: Antony Mayi
>
> When using mllib.recommendation.ALS the RDD returned by .predictAll() has all values pushed into single partition despite using quite high parallelism.
> This degrades performance of further processing (I can obviously run .partitionBy()) to balance it but that's still too costly (ie if running .predictAll() in loop for thousands of products) and should be possible to do it rather somehow on the model (automatically)).
> Bellow is an example on tiny sample (same on large dataset):
> {code:title=pyspark}
> >>> r1 = (1, 1, 1.0)
> >>> r2 = (1, 2, 2.0)
> >>> r3 = (2, 1, 2.0)
> >>> r4 = (2, 2, 2.0)
> >>> r5 = (3, 1, 1.0)
> >>> ratings = sc.parallelize([r1, r2, r3, r4, r5], 5)
> >>> ratings.getNumPartitions()
> 5
> >>> users = ratings.map(itemgetter(0)).distinct()
> >>> model = ALS.trainImplicit(ratings, 1, seed=10)
> >>> predictions_for_2 = model.predictAll(users.map(lambda u: (u, 2)))
> >>> predictions_for_2.glom().map(len).collect()
> [0, 0, 3, 0, 0]
> {code}



--
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