[
https://issues.apache.org/jira/browse/SPARK-9660?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14662303#comment-14662303
]
Feynman Liang edited comment on SPARK-9660 at 8/7/15 7:23 PM:
--------------------------------------------------------------
Logistic regression [only supports binary
classification|https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala#L83],
but various
[scaladocs|https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala#L50]
assert that this is a "backwards compatibility" feature, suggesting that
multiclass is supported.
This is made more confusing by the fact that inheriting from
{{ProbabilisticClassifier}} exposes a {{setThresholds(Array[Double])}} public
method, potentially allowing a user to set more than two thresholds on a binary
classifier... It may make sense to consider adding {{numClasses}} to
{{ClassifierParams}} and explicitly check that in {{HasThresholds}} (self type
annotation?)
was (Author: fliang):
Logistic regression [only supports binary
classification|https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala#L83],
but various
[scaladocs|https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala#L50]
assert that this is a "backwards compatibility" feature, suggesting that
multiclass is supported.
This is made more confusing by the fact that inheriting from
{{ProbabilisticClassifier}} exposes a {{setThresholds(Array[Double])}} public
method, potentially allowing a user to set more than two thresholds on a binary
classifier... It may make sense to consider adding {{numClasses}} to
{{ClassifierParams}} and explicitly check that when setting thresholds.
> ML 1.5 QA: API: New Scala APIs, docs
> ------------------------------------
>
> Key: SPARK-9660
> URL: https://issues.apache.org/jira/browse/SPARK-9660
> Project: Spark
> Issue Type: Sub-task
> Components: Documentation, ML, MLlib
> Reporter: Joseph K. Bradley
>
> Audit new public Scala APIs added to MLlib. Take note of:
> * Protected/public classes or methods. If access can be more private, then
> it should be.
> * Also look for non-sealed traits.
> * Documentation: Missing? Bad links or formatting?
> *Make sure to check the object doc!*
> As you find issues, please comment here, or better yet create JIRAs and link
> them.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
