[
https://issues.apache.org/jira/browse/MATH-1065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13895658#comment-13895658
]
Phil Steitz edited comment on MATH-1065 at 2/8/14 6:11 PM:
-----------------------------------------------------------
By contract in RealDistribution (same actually in IntegerDistribution), what
the inverse cum needs to return is
{code}
inf{x in R | P(X<=x) >= p} for 0 < p <= 1}
inf{x in R | P(X<=x) > 0} for p = 0.
{code}
Looks to me like your algorithm returns {code} sup{x in R | P(X <= x) < p}
{code} which is not the same. The key is to be consistent with the way we have
defined the inverse cum for both discrete ("Integer") and continuous ("Real")
distributions.
was (Author: psteitz):
By contract in RealDistribution (same actually in IntegerDistribution), what
the inverse cum needs to return is
inf{x in R | P(X<=x) >= p} for 0 < p <= 1}
inf{x in R | P(X<=x) > 0} for p = 0.
Looks to me like your algorithm returns sup{x in R | P(X <= x) < p} which is
not the same. The key is to be consistent with the way we have defined the
inverse cum for both discrete ("Integer") and continuous ("Real") distributions.
> EnumeratedRealDistribution.inverseCumulativeProbability returns values not in
> the samples set
> ---------------------------------------------------------------------------------------------
>
> Key: MATH-1065
> URL: https://issues.apache.org/jira/browse/MATH-1065
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.2
> Reporter: matteodg
> Fix For: 3.3
>
> Attachments: EnumeratedRealDistributionTest.java
>
>
> The method EnumeratedRealDistribution.inverseCumulativeProbability()
> sometimes returns values that are not in the initial samples domain...
> I will attach a test to exploit this bug.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
