[jira] [Commented] (GEOMETRY-21) Investigate Norm Accuracy

["Gilles (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/GEOMETRY-21?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16625298#comment-16625298 ] 

Gilles commented on GEOMETRY-21:
--------------------------------

{quote}failure
{quote}
Inserting some "print" statements:
{noformat}
0.10617094689696915 > 0.10617094689696706
0.10617094689696915 > 0.10617094689696706
failAssertCount=2 totalAssertCount=387
{noformat}
Thus, 2 failures (where the expectation is on the reversed inequality) that would pass if tolerance >= 1e-15 for a total of 387 similar inequality tests that pass.

{{testGeographicalMap()}} is a fairly big unit test, and letting the code continue past the above check completes successfully.

Commit be34ad93c0b0554ce5927811e0f762312172b9ea makes the test pass.
 Please have a look.

Also: magic numbers should be avoided (constant must be declared as a {{static final}} variable).

> Investigate Norm Accuracy
> -------------------------
>
>                 Key: GEOMETRY-21
>                 URL: https://issues.apache.org/jira/browse/GEOMETRY-21
>             Project: Apache Commons Geometry
>          Issue Type: Task
>            Reporter: Matt Juntunen
>            Priority: Minor
>
> Based on discussion in GEOMETRY-17, we should investigate the floating point accuracy of the current Vector normalization methods. Specifically, when the UnitVector private subclass in Vector3D is implemented to return exactly 1.0, the SphericalPolygonsSetTest#testGeographicalMap unit test in commons-geometry-enclosing begins to fail. We should
>  # Determine the cause of this failure.
>  # Determine if the current approach with the UnitVector subclasses introduces any issues with floating point accuracy.
>  # Add unit tests for Vector[123]D to quantify and verify the accuracy of the normalization.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)