[
https://issues.apache.org/jira/browse/SPARK-9312?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14642984#comment-14642984
]
Badari Madhav commented on SPARK-9312:
--------------------------------------
1. WRT OneVsRest I agree that the rawPrediction column should be interpreted as
"confidence" rather than probabilities but it is still very useful to determine
a threshold for the classifier. I will create a separate JIRA & PR combination
for OVR.
2. For the Logistic regression, I am planning to use it in a pipeline and I
guess it would be preferable to use the class under spark.ml. But I would think
it would be useful to have the api for pred prob in spark.mllib. Would that be
preferable or should I not bother making changes in
spark.mllib.LogisticRegressionModel?
> The multiclass logistic regression and the OneVsRest model do not provide
> predictive probability (confidence factor) along with the prediction
> ----------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-9312
> URL: https://issues.apache.org/jira/browse/SPARK-9312
> Project: Spark
> Issue Type: Improvement
> Components: ML, MLlib
> Affects Versions: 1.4.0, 1.4.1
> Reporter: Badari Madhav
> Labels: features
> Original Estimate: 72h
> Remaining Estimate: 72h
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
