spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Joseph K. Bradley (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-6158) Move private method boost in GradientBoostedTrees from Object to Class
Date Wed, 04 Mar 2015 18:48:38 GMT

    [ https://issues.apache.org/jira/browse/SPARK-6158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14347358#comment-14347358
] 

Joseph K. Bradley commented on SPARK-6158:
------------------------------------------

I'm not quite sure what the issue is.  It's often nice to use static methods (i.e., in objects)
with Spark to make sure we don't accidentally include a class-scope value in a closure (and
thus suck the entire class into a closure which must be sent to workers).  It can be a bit
confusing but just means you have to explicitly pass arguments to methods, rather than using
arguments defined at the class scope.

What do you mean by "I require the finally trained models in the private method boost"?  What
data do you need which you can't access?

> Move private method boost in GradientBoostedTrees from Object to Class
> ----------------------------------------------------------------------
>
>                 Key: SPARK-6158
>                 URL: https://issues.apache.org/jira/browse/SPARK-6158
>             Project: Spark
>          Issue Type: Improvement
>          Components: MLlib
>            Reporter: Manoj Kumar
>            Priority: Minor
>
> 1. a] When an instance of GradientBoostedTrees class is called and run, the boost method
in the GradientBoostedTrees object is called ultimately. b] When a GradientBoostedTrees object
is created and train is called, an instance of GradientBoostedTrees is created internally
and this calls the boost method in a newly instantiated GradientBoostedTrees object which
is confusing!
> 2. I require the finally trained models in the private method boost, for this (https://issues.apache.org/jira/browse/SPARK-6025)
(to prevent repeated training). Thus it makes sense to move boost to the class GradientBoostedTrees
where runWithValidation is already present.



--
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


Mime
View raw message