spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bryan Cutler (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (SPARK-10560) Make StreamingLogisticRegressionWithSGD Python API equals with Scala one
Date Tue, 06 Oct 2015 01:17:26 GMT

    [ https://issues.apache.org/jira/browse/SPARK-10560?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14944305#comment-14944305
] 

Bryan Cutler edited comment on SPARK-10560 at 10/6/15 1:17 AM:
---------------------------------------------------------------

Hi [~yanboliang], I just want to make sure I'm on the same page as to what we need to do here.
 
Here are the differences I see between the Python and Scala APIs for StreamingLogisticRegressionWithSGD:
-
* The documentation for Python is missing the default parameter values, also the same for
StreamingLinearRegressionWithSGD

* In Python StreamingLogisticRegressionWithSGD the regularization defaults to 0.01 while the
Scala version defaults to 0.  
I believe other SGD implementations default to non-zero, so maybe there is some reason to
turn it off in Streaming implementations?
In any case, these ones should probably default to the same value

* The Scala StreamingLogisticRegressionWithSGD is missing a method to set convergence tolerance,
it is in the Python one

* StreamingLogisticRegressionWithSGD for Scala and Python are missing ability to set regularization
parameter

* Python Streaming**RegressionWithSGD are missing API methods to set parameters, i.e. setStepSize

-
How about for this JIRA, I fix the documentation to include default parameters and then I
will make JIRAs for the other items?



was (Author: bryanc):
Hi [~yanboliang]], I just want to make sure I'm on the same page as to what we need to do
here.  
Here are the differences I see between the Python and Scala APIs for StreamingLogisticRegressionWithSGD:
-
* The documentation for Python is missing the default parameter values, also the same for
StreamingLinearRegressionWithSGD

* In Python StreamingLogisticRegressionWithSGD the regularization defaults to 0.01 while the
Scala version defaults to 0.  
I believe other SGD implementations default to non-zero, so maybe there is some reason to
turn it off in Streaming implementations?
In any case, these ones should probably default to the same value

* The Scala StreamingLogisticRegressionWithSGD is missing a method to set convergence tolerance,
it is in the Python one

* StreamingLogisticRegressionWithSGD for Scala and Python are missing ability to set regularization
parameter

* Python Streaming**RegressionWithSGD are missing API methods to set parameters, i.e. setStepSize

How about for this JIRA, I fix the documentation to include default parameters and then I
will make JIRAs for the other items?


> Make StreamingLogisticRegressionWithSGD Python API equals with Scala one
> ------------------------------------------------------------------------
>
>                 Key: SPARK-10560
>                 URL: https://issues.apache.org/jira/browse/SPARK-10560
>             Project: Spark
>          Issue Type: Sub-task
>          Components: MLlib, PySpark
>            Reporter: Yanbo Liang
>            Priority: Minor
>
> StreamingLogisticRegressionWithSGD Python API lacks of some parameters compared with
Scala one, here we make them equality.



--
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