Dear Bill,
Thank you for the GIF picture and your comments. As you mentioned the set of
3D points was rather special. Points are scattered but... ...most of them
are co-planar and co-linear. We were expecting some NaN as a result because
we perform interpolation on the whole cube (we took the Min and Max of each
direction, x, y and z to set the limit of our interpolation domain) while
the points only occupy of portion of it.
Regards,
Sebastien Lechelle
----- Original Message -----
From: Bill Hibbard <hibbard@xxxxxxxxxxxxxxxxx>
To: <visad-list@xxxxxxxxxxxxx>
Sent: Monday, May 07, 2001 7:03 PM
Subject: Re: Interpolation returning NaN...
> I just want to add a little to what Tom said. His new version
> of DataFitTest3dGridloss.java (in dataFit.jar) produces a very
> nice visualization of the situation. I have attached a GIF
> snapshot.
>
> As Tom says, this GIF shows that some points lie outside of the
> convex hull of the Irregular3DSet, so we expect NaNs from the
> interpolation.
>
> The GIF also shows a lot of regularity in the Irregular3DSet,
> especially groups of points lying on the same planes and the
> same lines. Co-located, co-linear and co-planar points always
> give Deluany algorithms trouble. In this example, the co-planar
> points generate some tetrahedra with zero volume, and I think
> also generate ambiguities in tetrahedralizations.
>
> Tom fixed a bug in Irregular3DSet that manifested in the case
> of zero-volume tetrahedra. There may still be disagreements
> with interpolations from another system (e.g., Matlab) due to
> the ambiguity of tetrahedralization.
>
> Thank for bringing this to our attention, Sebastion.
>
> Cheers,
> Bill
>
> > I've implemented and tested a fix to the problem you discovered with
> > your interpolation testing - sorry for the delay. Herein you'll find
> > two attachments: Irregular3DSet.jar, dataFit.jar. Apply jar xvf to
> > the first of these in your visad directory. Jar xvf the second in the
> > top level of your classpath; note that this jar file contains my
> > modifications of the source you were kind enough to provide. It will
> > display the dataset, as well as, any NaN points which will be blue.
> > You should see the blue (NaN interpolants) outside the hull of your
> > dataset.
> >
> > Let me know if have any problems,
> >
> > TomR
> >
>
--------------------------------------------------------------------------
----------------------------------------------------------------------
> > Name: Irregular3DSet.jar
> > Irregular3DSet.jar Type: Java Archive
(application/java-archive)
> > Encoding: base64
> > Download Status: Not downloaded with message
> >
> > Name: dataFit.jar
> > dataFit.jar Type: Java Archive (application/java-archive)
> > Encoding: base64
> > Download Status: Not downloaded with message
>
> --
> ----------------------------------------------------------
> Bill Hibbard, SSEC, 1225 W. Dayton St., Madison, WI 53706
> hibbard@xxxxxxxxxxxxxxxxx 608-263-4427 fax: 608-263-6738
> http://www.ssec.wisc.edu/~billh/vis.html
----------------------------------------------------------------------------
----
