[jira] [Reopened] (MATH-1208) EmpiricalDistribution cumulativeProbability can return NaN when evaluated within a constant bin

[Phil Steitz (JIRA)]

     [ 
https://issues.apache.org/jira/browse/MATH-1208?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Phil Steitz reopened MATH-1208:
-------------------------------

I think the fix I committed in master: b148046add84f6c5bd19aeceb10c1a0fcf2e1283 
is incorrect.  When a constant kernel is explicitly set, behavior should pass 
the original version of testKernelOverrideConstant (with within-bin values less 
than the point mass not contributing the mass of the bin to the cumulative 
probability).   The special handling in cumulativeProbability should be
{code}
if (x < kernel.getNumericalMean()) {
    return pBminus;
 } else {
     return pBminus + pB;
 }
{code}


> EmpiricalDistribution cumulativeProbability can return NaN when evaluated 
> within a constant bin
> -----------------------------------------------------------------------------------------------
>
>                 Key: MATH-1208
>                 URL: https://issues.apache.org/jira/browse/MATH-1208
>             Project: Commons Math
>          Issue Type: Bug
>    Affects Versions: 3.4.1
>            Reporter: Phil Steitz
>             Fix For: 4.0, 3.5
>
>
> If x belongs to a bin with no variance or to which a ConstantRealDistribution 
> kernel has been assigned, cumulativeProbability(x) can return NaN.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)