[
https://issues.apache.org/jira/browse/GEOMETRY-71?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17018601#comment-17018601
]
Matt Juntunen commented on GEOMETRY-71:
---------------------------------------
I just added a follow-up pull request:
https://github.com/apache/commons-geometry/pull/57.
I was able to remove nearly all usages of the {{BARYCENTER_EPS}} value in the
tests, with two exceptions in {{RegionBSPTree2STest}}: one in the computation
of the barycenter of France and the other in the circle test with the very
small radius (0.0001). However, I was able to lower the epsilon value from 1e-2
to 1e-5 so we have a considerably higher level of precision now.
I also updated the barycenter computation in {{RegionBSPTree1S}} to use the
same weighted-barycenter approach as {{RegionBSPTree2S}}. They both now return
null when no _unique_ barycenter exists.
Any objections to closing this issue after this PR is merged?
> Investigate Spherical Barycenter Accuracy
> -----------------------------------------
>
> Key: GEOMETRY-71
> URL: https://issues.apache.org/jira/browse/GEOMETRY-71
> Project: Apache Commons Geometry
> Issue Type: Bug
> Reporter: Matt Juntunen
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.0
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> The current code for computing spherical barycenters in
> {{ConvexArea2S.getBarycenter()}} seems to suffer from floating point accuracy
> issues. The {{ConvexArea2STest.checkBarycenterConsistency()}} method checks
> the consistency of the barycenter computation of a region by splitting the
> region into two sections, computing the area and barycenter of each section,
> and then computing the combined barycenter of the sections by adding the
> barycenter of each scaled by its corresponding area. It is expected that the
> combined barycenter computed in this way should equal the barycenter computed
> for the region as a whole. However, in practice, a large epsilon value is
> needed in the comparison in order for the tests to pass. We need to
> investigate why this is the case.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
