|
|
3.4.3 Calling the Covariance Function by Name or by Function
(cov.by.name logical parameter)
What in the world does this parameter do?
cov.by.name is a
logical parameter (default = T) that tells Krig whether to call the
covariance function (cov.function)
indirectly by name or by function. If cov.by.name is true, then it will use the
standard S function do.call to evaluate
the function. The optional arguments specific to the particular
covariance function are stored in a list object called "args" that
holds all other pertinent information. This is convenient for passing
parameters from one function to another. For example, instead of
finding the covariance matrix, k(x1, x2), by using the function
directly--i.e. cov.function( x1, x2,
...)--the function evaluation would proceed by:
do.call( out$call.name, c( out$args, list( x1, x2)))
may appear to be more complicated, but more
compatible with the implementation of the S language in the R
statistical package. Note that out$call.name is a character value that simply
stores the name of the covariance function to be called. For example,
if the default exp.cov is used, then
out$call.name is "exp.cov".
If cov.by.name is flase this results in a copy of the covariance function,
with the correct arguments attached to the output Krig
object. The advantage is that the data, estimate and specific
covariance S-function are bundled together in a single object. Even if the
covariance function is inadvertently deleted or modified elsewhere, using the
returned Krig object will still give the correct
results. The disadvantage is that the covariance S-function is always copied
and is therefore wasteful of storage. Also, this copying modification of the function is problematic in the R-language.
|