# [R] smooth.spline

Prof Brian Ripley ripley at stats.ox.ac.uk
Mon Apr 17 17:20:40 CEST 2006

I believe this is covered by

The \code{x} vector should contain at least four distinct values.
\emph{Distinct} here means \dQuote{distinct after rounding to 6 significant
digits}, i.e., \code{x} will be transformed to
\code{unique(sort(signif(x, 6)))}, and \code{y} and \code{w} are
pooled accordingly.

but without a working example it is hard to tell.

On Fri, 17 Mar 2006, Rolf Turner wrote:

>
> I have noticed a slightly puzzling behaviour exhibited by
> smooth.spline().  If I do
>
> 	sss <- smooth.spline(x,y)
>
> for a certain pair of data vectors x and y, and then do
>
> 	length(sss\$x)
>
> I get the result 18''.  However if I do
>
> 	length(unique(x))
>
> I get 27''.  Trying to force smooth.spline() to use more knots I
> tried
>
> 	sss <- smooth.spline(x,y,all.knots=TRUE)
>
> but again got output of length 18 again.  I then tried
>
> 	sss <- smooth.spline(x,y,nknots=27)
>
> and got an error message.  Finally I was inspired to re-scale the
> x-values.  Doing
>
> 	x <- x - min(x)
> 	sss <- smooth.spline(x,y)
>
> gave me output of length 27 as expected.  Apparently smooth.spline()
> is using a criterion for distinctness of x values which is based on a
> tolerance relative to the size of the x's.  (I.e. a different
> criterion from unique().)  While this is assuredly a Good Thing, it
> might be nice to have some sort of warning in the help file to keep
> young players from getting confused.

I think there is one, if this is the phenomenon I know about.

--
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595