You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@commons.apache.org by "Gal Lalouche (JIRA)" <ji...@apache.org> on 2013/12/05 21:37:36 UTC

[jira] [Reopened] (MATH-1068) KendallsCorrelation suffers from integer overflow for large arrays.

     [ https://issues.apache.org/jira/browse/MATH-1068?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Gal Lalouche reopened MATH-1068:
--------------------------------


I've noticed a few more overflow issues for very large arrays (100k's), e.g., correlation < -1 or > 1.
Changing all the tiedX/Y/XYPairs, consecutiveX/Y/XYTies and swaps to long fixed this.

> KendallsCorrelation suffers from integer overflow for large arrays.
> -------------------------------------------------------------------
>
>                 Key: MATH-1068
>                 URL: https://issues.apache.org/jira/browse/MATH-1068
>             Project: Commons Math
>          Issue Type: Bug
>    Affects Versions: 3.3
>            Reporter: Gal Lalouche
>            Priority: Minor
>              Labels: newbie
>             Fix For: 3.3
>
>   Original Estimate: 1m
>  Remaining Estimate: 1m
>
> For large array size (say, over 5,000), numPairs > 10 million.
> in line 258, (numPairs - tiedXPairs) * (numPairs - tiedYPairs) possibly > 100 billion, which will cause an integer overflow, resulting in a negative number, which will result in the end result in a NaN since the square-root of that number is calculated.
> This can easily be solved by changing line 163 to
> final long numPairs = ((long)n) * (n - 1) / 2; // to avoid overflow



--
This message was sent by Atlassian JIRA
(v6.1#6144)