[
https://issues.apache.org/jira/browse/MATH-1490?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16914937#comment-16914937
]
Gilles commented on MATH-1490:
------------------------------
I've done it. So:
{noformat}
$ git diff MATH_3_3 MATH_3_4 --
src/main/java/org/apache/commons/math3/stat/descriptive/rank/Percentile.java
{noformat}
displays a large number of changes between v3.3 and v3.4 (in order to add new
features).
I'm not too surprised that this could entail a 1e-15 relative change of the
output. Moreover, the unit tests show an expected level of precision much lower
than that (although I admit that it's not clear why).
{quote}it breaks my existing code. I got an incompatible error.
{quote}
Could you elaborate a little?
> Percentile computational accuracy issue
> ---------------------------------------
>
> Key: MATH-1490
> URL: https://issues.apache.org/jira/browse/MATH-1490
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.4, 3.4.1, 3.5, 3.6, 3.6.1
> Environment: System: Linux testinglab 4.4.0-131-generic
> #157~14.04.1-Ubuntu
> Java version "1.8.0_191"
> Java(TM) SE Runtime Environment (build 1.8.0_191-b12)
> Java HotSpot(TM) 64-Bit Server VM (build 25.191-b12, mixed mode)
>
>
> Reporter: xia0c
> Assignee: Rob Tompkins
> Priority: Major
> Labels: performance
> Attachments: BugDemo.java
>
>
> The percentile method works well on the older versions, e.g., the version
> before 3.4. However, when I update commons-math to the newer version, there
> produces a computational accuracy issue. There is a backward compatibility
> bug behind it.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
