[
https://issues.apache.org/jira/browse/SPARK-25309?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ravi updated SPARK-25309:
-------------------------
Description: SPARK-19357 and SPARK-21911 have helped parallelize Pipelines
in Spark. However, instead of setting the parallelism Parameter in the
CrossValidator it would be good to have something like njobs=-1 (like Scikit
Learn) where the Pipeline DAG could be automatically parallelized and scheduled
based on the resources allocated to the Spark Session instead of having the
user pick the integer value for this parameter. (was: SPARK-19357 and
SPARK-21911 have helped parallelize Pipelines in Spark. However, instead of
setting the parallelism Parameter in the CrossValidator it would be good to
have something like njobs=-1 (like Scikit Learn) where the Pipleline DAG could
be automatically parallelized and scheduled based on the resources allocated to
the Spark Session instead of having the user pick the integer value for this
parameter. )
> Sci-kit Learn like Auto Pipeline Parallelization in Spark
> ----------------------------------------------------------
>
> Key: SPARK-25309
> URL: https://issues.apache.org/jira/browse/SPARK-25309
> Project: Spark
> Issue Type: Improvement
> Components: PySpark
> Affects Versions: 2.3.1
> Reporter: Ravi
> Priority: Critical
>
> SPARK-19357 and SPARK-21911 have helped parallelize Pipelines in Spark.
> However, instead of setting the parallelism Parameter in the CrossValidator
> it would be good to have something like njobs=-1 (like Scikit Learn) where
> the Pipeline DAG could be automatically parallelized and scheduled based on
> the resources allocated to the Spark Session instead of having the user pick
> the integer value for this parameter.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
