The R Project SVN R

Rev

Rev 72404 | Rev 81583 | Go to most recent revision | Show entire file | Ignore whitespace | Details | Blame | Last modification | View Log | RSS feed

Rev 72404 Rev 77213
Line 163... Line 163...
163 
  internal code uses an approximation that involves leaving out groups
 163 
  internal code uses an approximation that involves leaving out groups
164 
  of duplicated points.  \code{cv = TRUE} is best avoided in that case.
 164 
  of duplicated points.  \code{cv = TRUE} is best avoided in that case.
165 
}
 165 
}
166 
\note{
 166 
\note{
167 
  The number of unique \code{x} values, \eqn{\code{nx} = n_x}{nx}, are
 167 
  The number of unique \code{x} values, \eqn{\code{nx} = n_x}{nx}, are
168 
  determined by the \code{tol} argument, equivalently to \preformatted{%
 168 
  determined by the \code{tol} argument, equivalently to
- 
 
 169 
  \preformatted{
169 
    nx <- length(x)  -  sum(duplicated( round((x - mean(x)) / tol) ))}
 170 
    nx <- length(x) - sum(duplicated( round((x - mean(x)) / tol) ))
- 
 
 171 
  }
170 
 
 172 
 
171 
  The default \code{all.knots = FALSE} and \code{nknots = .nknots.smspl},
 173 
  The default \code{all.knots = FALSE} and \code{nknots = .nknots.smspl},
172 
  entails using only \eqn{O({n_x}^{0.2})}{O(nx ^ 0.2)}
 174 
  entails using only \eqn{O({n_x}^{0.2})}{O(nx ^ 0.2)}
173 
  knots instead of \eqn{n_x}{nx} for \eqn{n_x > 49}{nx > 49}.  This cuts
 175 
  knots instead of \eqn{n_x}{nx} for \eqn{n_x > 49}{nx > 49}.  This cuts
174 
  speed and memory requirements, but not drastically anymore since \R
 176 
  speed and memory requirements, but not drastically anymore since \R