[jira] [Commented] (SPARK-12806) Support SQL expressions extracting values from VectorUDT

["John Bauer (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-12806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16964475#comment-16964475 ] 

John Bauer commented on SPARK-12806:
------------------------------------

This is still a problem.  For example, classification models emit probability as a VectorUDT, which are unusable in PySpark.  This makes constructing boosting/bagging algorithms or even just using them as additional features in a second model problematic.

> Support SQL expressions extracting values from VectorUDT
> --------------------------------------------------------
>
>                 Key: SPARK-12806
>                 URL: https://issues.apache.org/jira/browse/SPARK-12806
>             Project: Spark
>          Issue Type: Improvement
>          Components: MLlib, SQL
>    Affects Versions: 1.6.0
>            Reporter: Feynman Liang
>            Priority: Major
>              Labels: bulk-closed
>
> Use cases exist where a specific index within a {{VectorUDT}} column of a {{DataFrame}} is required. For example, we may be interested in extracting a specific class probability from the {{probabilityCol}} of a {{LogisticRegression}} to compute losses. However, if {{probability}} is a column of {{df}} with type {{VectorUDT}}, the following code fails:
> {code}
> df.select("probability.0")
> AnalysisException: u"Can't extract value from probability"
> {code}
> thrown from {{sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypeExtractors.scala}}.
> {{VectorUDT}} essentially wraps a {{StructType}}, hence one would expect it to support value extraction Expressions in an analogous way.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org