[
https://issues.apache.org/jira/browse/SPARK-11939?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15034580#comment-15034580
]
Joseph K. Bradley commented on SPARK-11939:
-------------------------------------------
* save for Estimator/Transformer: I agree.
* load Transformer(Model): I agree it is doable. We technically do not need to
make the constructors public since Python ignores private/public. I'd prefer
that.
* load for Estimator: For most Estimators, we could probably have a default
implementation which loads the Java object, creates the empty Python Estimator,
and then copies all Params it can find. That should be doable with a single
implementation. I don't think we need to modify the constructors.
** For complex Estimators with Params of complex types, we may need a
specialized implementation in Python. I hope that we can still use the Java
save() implementation for these to make sure all languages use the same format.
> PySpark support model export/import for Pipeline API
> ----------------------------------------------------
>
> Key: SPARK-11939
> URL: https://issues.apache.org/jira/browse/SPARK-11939
> Project: Spark
> Issue Type: Sub-task
> Components: ML, PySpark
> Reporter: Yanbo Liang
>
> SPARK-6725 provide model export/import for Pipeline API at Scala side, we
> should also support it at Python side.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
