Hi Bill, Thank you for the warning below. Here is a shorter copy of my previous mail with some Java files attached to it (in a zip format). Regards, Sebastien >Hi Sebastien, > >I'm afraid our dumb majordomo bounced your message for being >too long. Can you please resend, and put any big files on a >server with just the URL in the message. > >Thanks, >Bill >---------------------------------------------------------- >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 Dear all, I don't know if I am asking too much, but maybe someone from the mailing-list will be able to help me... I work on a Java software project where we need to do some 2D and 3D-interpolation computation. Since we are not expert in Geometrical computing we decided to use the excellent VisaD API (no need to try to duplicate what has already been done!!!). To facilitate the work of my colleagues, I also developed some wrapper classes of the interpolation functions of VisaD. These 2 classes Interpolation2DLinear and Interpolation3DLinear are available below with two other classes (if someone is interested). The classes are fairly easy to use... One constructs a function v[] = F(x[],y[],z[]) like this : Interpolation3DLinear my_interp = new Interpolation3DLinear(x[],y[],z[],v[]) One computes the interpolation at (xi,yi,zi) like this : my_interp.evaluate(xi,yi,zi) All data are arrays of double and double (see source of the classes). Data are scaled before interpolation to avoid unit problems... So we recently tested the wrapper and did some comparison with the interpolation functionality of Matlab 6. For small function (i.e. functions defined by a small sample of scattered points), the results were identical and I was really happy about that... until one of my colleague wrote a test file DataFitTest3dGridloss.java (available below) which tests Interpolation3DLinear with a function sampled on a lot of points... The test file performs interpolation on 168 points of the domain and all of them except 1 returns NaN. We compared the only numerical result from Interpolation3DLinear with Matlab and the values are different... Does anyone knows about such a phenomenon? Is there a limit in the number of points a Irregular3DSet can take??? Is there any geometrical restriction on the set of points for Irregular3DSet ??? Am I doing something wrong (I suppose so, but I can't find what it is)... I would grateful if someone could help me... ...propose some possible explanations ...or even just reproduce the problem... Regards, Sebastien
Attachment:
dataFit.zip
Description: Zip compressed data
visad archives: