spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Peter Rudenko (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-7131) Move tree,forest implementation from spark.mllib to spark.ml
Date Wed, 09 Dec 2015 20:52:10 GMT

    [ https://issues.apache.org/jira/browse/SPARK-7131?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15049368#comment-15049368
] 

Peter Rudenko commented on SPARK-7131:
--------------------------------------

Please remove final classes from RF and GBM models in ml package. I want to extend them, set
some parameters, reimplement some functionality (do probabilistic models for GBC, etc.).

> Move tree,forest implementation from spark.mllib to spark.ml
> ------------------------------------------------------------
>
>                 Key: SPARK-7131
>                 URL: https://issues.apache.org/jira/browse/SPARK-7131
>             Project: Spark
>          Issue Type: Improvement
>          Components: ML, MLlib
>    Affects Versions: 1.4.0
>            Reporter: Joseph K. Bradley
>            Assignee: Joseph K. Bradley
>             Fix For: 1.5.0
>
>   Original Estimate: 168h
>  Remaining Estimate: 168h
>
> We want to change and improve the spark.ml API for trees and ensembles, but we cannot
change the old API in spark.mllib.  To support the changes we want to make, we should move
the implementation from spark.mllib to spark.ml.  We will generalize and modify it, but will
also ensure that we do not change the behavior of the old API.
> There are several steps to this:
> 1. Copy the implementation over to spark.ml and change the spark.ml classes to use that
implementation, rather than calling the spark.mllib implementation.  The current spark.ml
tests will ensure that the 2 implementations learn exactly the same models.  Note: This should
include performance testing to make sure the updated code does not have any regressions. -->
*UPDATE*: I have run tests using spark-perf, and there were no regressions.
> 2. Remove the spark.mllib implementation, and make the spark.mllib APIs wrappers around
the spark.ml implementation.  The spark.ml tests will again ensure that we do not change any
behavior.
> 3. Move the unit tests to spark.ml, and change the spark.mllib unit tests to verify model
equivalence.
> This JIRA is now for step 1 only.  Steps 2 and 3 will be in separate JIRAs.
> After these updates, we can more safely generalize and improve the spark.ml implementation.



--
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