[jira] [Commented] (MATH-1361) "NaturalRankingTest" broken?

["Phil Steitz (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MATH-1361?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16215563#comment-16215563 ] 

Phil Steitz commented on MATH-1361:
-----------------------------------

This is the result of sloppy test implementation (by me).  The TiesRandom strategy assigns ranks randomly among the applicable ranks, so the expected values are in fact stochastic.  To eliminate seed dependency, the expected values for ties need to be allowed to vary over the possible ranks.  One way to do that is the fix that I applied to Hipparchus for this issue [https://github.com/Hipparchus-Math/hipparchus/commit/9cc199a7c7b2051a02bd1db060caf43a00748f2b].

> "NaturalRankingTest" broken?
> ----------------------------
>
>                 Key: MATH-1361
>                 URL: https://issues.apache.org/jira/browse/MATH-1361
>             Project: Commons Math
>          Issue Type: Bug
>    Affects Versions: 3.6.1
>            Reporter: Gilles
>              Labels: test
>             Fix For: 4.0
>
>
> As reported on the ["dev" ML|http://markmail.org/message/2oqkvx33arjj34lz], the test method
> {noformat}
> testNaNsFixedTiesRandom()
> {noformat}
> reports failure for more than 96 out of 100 seed values.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)