You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Sean Owen (JIRA)" <ji...@apache.org> on 2017/08/30 10:37:00 UTC
[jira] [Resolved] (SPARK-21806) BinaryClassificationMetrics pr():
first point (0.0, 1.0) is misleading
[ https://issues.apache.org/jira/browse/SPARK-21806?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sean Owen resolved SPARK-21806.
-------------------------------
Resolution: Fixed
Fix Version/s: 2.3.0
Issue resolved by pull request 19038
[https://github.com/apache/spark/pull/19038]
> BinaryClassificationMetrics pr(): first point (0.0, 1.0) is misleading
> ----------------------------------------------------------------------
>
> Key: SPARK-21806
> URL: https://issues.apache.org/jira/browse/SPARK-21806
> Project: Spark
> Issue Type: Improvement
> Components: MLlib
> Affects Versions: 2.2.0
> Reporter: Marc Kaminski
> Assignee: Sean Owen
> Priority: Minor
> Labels: releasenotes
> Fix For: 2.3.0
>
> Attachments: PRROC_example.jpeg
>
>
> I would like to reference to a [discussion in scikit-learn| https://github.com/scikit-learn/scikit-learn/issues/4223], as this behavior is probably based on the scikit implementation.
> Summary:
> Currently, the y-axis intercept of the precision recall curve is set to (0.0, 1.0). This behavior is not ideal in certain edge cases (see example below) and can also have an impact on cross validation, when optimization metric is set to "areaUnderPR".
> Please consider [blucena's post|https://github.com/scikit-learn/scikit-learn/issues/4223#issuecomment-215273613] for possible alternatives.
> Edge case example:
> Consider a bad classifier, that assigns a high probability to all samples. A possible output might look like this:
> ||Real label || Score ||
> |1.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 1.0 |
> |0.0 | 0.95 |
> |0.0 | 0.95 |
> |1.0 | 1.0 |
> This results in the following pr points (first line set by default):
> ||Threshold || Recall ||Precision ||
> |1.0 | 0.0 | 1.0 |
> |0.95| 1.0 | 0.2 |
> |0.0| 1.0 | 0,16 |
> The auPRC would be around 0.6. Classifiers with a more differentiated probability assignment will be falsely assumed to perform worse in regard to this auPRC.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org