nychka at cgd.ucar.edu
Thu May 16 18:58:31 CEST 2002
Part of fitting a thin plate spline involves setting the smoothing
parameter (usually referred to as lambda). This is very similar to
estimating the nugget variance in Kriging. At the extreme values (0 or
infinity) of this parameter the spline either is just a low order
polynomial function fit by least squares( lambda= infinity) or is a
smooth surface that passes through every data point ( lambda= 0).
The first message indicates that for the data fit below one of these
choices is suggested by cross-validation. The summary of the Tps object
will tell you which one; look at the effective number of parameters for
The next thing to try is to plot the returned Tps object and examine the
GCV curve. Often it is very flat and many values of lambda are close to
The second message is the same using a slightly different version of GCV.
The third message comes in trying to match the smoothing parameter to an
estimate of pure error derived from replicated points. This message
indicates that there is no value of the smoothing parameter that gives an
spline-based estimate of sigma that will match the estimate of sigma based
on replicated data. That is not always bad because the pure error estimate
can be quite variable.
Hope this helps,
Rachid send me mail privately if you have more questions.
Geophysical Statistics Project Email: nychka at ucar.edu
National Center for Atmospheric Research Voice: 303-497-1711
PO Box 3000 FAX: 303-497-1333
Boulder, CO 80307-3000 Web: www.cgd.ucar.edu/~nychka
Address for overnight mail: 1850 Table Mesa Drive, Boulder, CO 80305
On Thu, 16 May 2002, Rachid Cheddadi wrote:
> I have a 4 column file (long/lat/elev/variable) and I tried to fit the
> values of my variable to the XYZ space using Tps and I keep getting the
> following message:
> Warning messages:
> 1: GCV search gives a minumum at the endpoints of the grid search in:
> Krig.find.gcvmin(info, lambda.grid, gcv.grid$GCV, Krig.fgcv,
> 2: GCV search gives a minumum at the endpoints of the grid search in:
> Krig.find.gcvmin(info, lambda.grid, gcv.grid$GCV.one, Krig.fgcv.one,
> 3: Failed to find the upcrossing in: find.upcross(Krig.fs2hat, info,
> upcross.level = shat.pure.error^2,
> Any help please?
> Many thanks
> Dr. Rachid Cheddadi
> European Pollen Database Tel: 00.33.(0)126.96.36.199.18
> Ancien Archeveche Fax: 00.33.(0)188.8.131.52.03
> 13200 Arles -France email: rachid.cheddadi at lbhp.u-3mrs.fr
> r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
> Send "info", "help", or "[un]subscribe"
> (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch
More information about the R-help