The function knn.forecast.boot.intervals() has been
added to provide the capability to generate interval forecasts and
simulations with the package.
The default value of max.k in
knn.forecast.randomsearch.tuning() has been changed from
NA to NULL, to bring it in line with the
approach I took to optional arguments elsewhere in the package. However,
passing max.k = NA will still work, and behave in the same
manner as max.k = NULL.
Changing the behavior of max.k in
knn.forecast.randomsearch.tuning() when the current or
previous default argument is passed to be more dynamic based on the
length of the input series. Now max.k will be set to
min(floor((length(y.in)) * .4), length(y.in) - val.holdout.len - test.h)
if NULL or NA is passed. This change was made
because the more arbitrary behavior in knnwtsim 0.1.0 where
max.k would be set to
min(floor((length(y.in)) * .4), 50) if NA was
passed could lead to errors.
Error handling to throw errors or warnings for conflicting
arguments, arguments outside reasonable ranges, and argument types
differing from those listed in the help files has been added to all user
facing functions in the package: StMatrixCalc(),
SpMatrixCalc(), SxMatrixCalc(),
SwMatrixCalc(), knn.forecast(),
knn.forecast.randomsearch.tuning(), and
knn.forecast.boot.intervals().
A min.k argument has been added to
knn.forecast.randomsearch.tuning(), which can be used set a
floor for the minimum number of nearest neighbors to be proposed in any
parameter set to be tested in tuning. By default min.k = 1,
in line with previous behavior of the function.
Added references to arXiv:2112.06266 to DESCRIPTION
and help files as needed to provide more information on
methodology.