[ 
https://issues.apache.org/jira/browse/MATH-891?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13615500#comment-13615500
 ] 

Phil Steitz commented on MATH-891:
----------------------------------

I am OK with committing this patch, but lets keep the issue open, or open 
another one for handling missing data in multivariate stats.  I think it is OK 
to leave the RankingAlgorithm interface and impls as is - they are doing what 
they should be doing by contract.  I think multivariate stats should just not 
allow the REMOVED NAN strategy (i.e. throw in this case).  Once we agree on how 
to implement and represent missing data strategies at least just for this 
class, the Spearman's constructor should then be modified to include 
specification of missing data strategy.

I think it is better to commit the workaround now, since behavior is currently 
broken; but note in the javadoc and release notes that as of 4.0, the 
constructor will throw on REMOVED NaNStrategy and NANs should not be used to 
represent missing data.  Practical advice to users is to preprocess data to 
remove / replace / impute missing data in preparation for this.
                
> SpearmansCorrelation fails when using NaturalRanking together with 
> NaNStrategy.REMOVED
> --------------------------------------------------------------------------------------
>
>                 Key: MATH-891
>                 URL: https://issues.apache.org/jira/browse/MATH-891
>             Project: Commons Math
>          Issue Type: Bug
>    Affects Versions: 3.0
>            Reporter: Thomas Neidhart
>         Attachments: MATH-891.patch
>
>
> As reported by Martin Rosellen on the users mailinglist:
> Using a NaturalRanking with a REMOVED NaNStrategy can result in an exception 
> when NaN are contained in the input arrays.
> The current implementation just removes the NaN values where they occur, 
> without taken care to remove the corresponding values in the other array.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to