[jira] [Commented] (SPARK-16470) ml.regression.LinearRegression training data do not check whether the result actually reach convergence

["Apache Spark (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-16470?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15369441#comment-15369441 ] 

Apache Spark commented on SPARK-16470:
--------------------------------------

User 'WeichenXu123' has created a pull request for this issue:
https://github.com/apache/spark/pull/14122

> ml.regression.LinearRegression training data do not check whether the result actually reach convergence
> -------------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-16470
>                 URL: https://issues.apache.org/jira/browse/SPARK-16470
>             Project: Spark
>          Issue Type: Improvement
>          Components: ML, Optimizer
>    Affects Versions: 2.0.1, 2.1.0
>            Reporter: Weichen Xu
>   Original Estimate: 24h
>  Remaining Estimate: 24h
>
> In `ml.regression.LinearRegression`, it use breeze `LBFGS` and `OWLQN` optimizer to do data training,  but do not check whether breeze's optimizer returned result  actually reached convergence.
> In fact, the  `LBFGS` and `OWLQN`  optimizer in breeze finish iteration may result the following situations:
> 1) reach max iteration number
> 2) function reach value convergence
> 3) objective function stop improving
> 4) gradient reach convergence
> 5) search failed(due to some internal numerical error)
> only (2) & (4) situations above can be regarded as convergence result.
> in other situation, we should at least print some warning to remind spark user.



--
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