Get Multivariate Functional Data from a data frame
get_mfd_df(
dt,
domain,
arg,
id,
variables,
n_basis = 30,
n_order = 4,
basisobj = NULL,
Lfdobj = 2,
lambda = NULL,
lambda_grid = 10^seq(-10, 1, length.out = 10),
ncores = 1
)
A data.frame
containing the discrete data.
For each functional variable, a single column,
whose name is provided in the argument variables
,
contains discrete values of that variable for all functional observation.
The column indicated by the argument id
denotes which is the functional observation in each row.
The column indicated by the argument arg
gives the argument value at which
the discrete values of the functional variables are observed for each row.
A numeric vector of length 2 defining the interval over which the functional data object can be evaluated.
A character variable, which is the name of
the column of the data frame dt
giving the argument values at which the functional variables
are evaluated for each row.
A character variable indicating which is the functional observation in each row.
A vector of characters of the column names
of the data frame dt
indicating the functional variables.
An integer variable specifying the number of basis functions; default value is 30. See details on basis functions.
An integer specifying the order of b-splines, which is one higher than their degree. The default of 4 gives cubic splines.
An object of class basisfd
defining
the basis function expansion.
Default is NULL
, which means that
a basisfd
object is created by doing
create.bspline.basis(rangeval = domain,
nbasis = n_basis, norder = n_order)
An object of class Lfd
defining a
linear differential operator of order m.
It is used to specify a roughness penalty through fdPar
.
Alternatively, a nonnegative integer
specifying the order m can be given and is
passed as Lfdobj
argument to the function fdPar
,
which indicates that the derivative of order m is penalized.
Default value is 2, which means that the
integrated squared second derivative is penalized.
A non-negative real number.
If you want to use a single specified smoothing parameter
for all functional data objects in the dataset,
this argument is passed to the function fda::fdPar
.
Default value is NULL, in this case the smoothing parameter is chosen
by minimizing the generalized cross-validation (GCV)
criterion over the grid of values given by the argument.
See details on how smoothing parameters work.
A vector of non-negative real numbers.
If lambda
is provided as a single number, this argument is ignored.
If lambda
is NULL, then this provides the grid of values
over which the optimal smoothing parameter is
searched. Default value is 10^seq(-10,1,l=20)
.
If you want parallelization, give the number of cores/threads to be used when doing GCV separately on all observations.
An object of class mfd
.
See also ?mfd
for additional details on the
multivariate functional data class.
Basis functions are created with
fda::create.bspline.basis(domain, n_basis)
, i.e.
B-spline basis functions of order 4 with equally spaced knots
are used to create mfd
objects.
The smoothing penalty lambda is provided as
fda::fdPar(bs, 2, lambda)
,
where bs is the basis object and 2 indicates
that the integrated squared second derivative is penalized.
Rather than having a data frame with long format,
i.e. with all functional observations in a single column
for each functional variable,
if all functional observations are observed on a common equally spaced grid,
discrete data may be available in matrix form for each functional variable.
In this case, see get_mfd_list
.
library(funcharts)
x <- seq(1, 10, length = 25)
y11 <- cos(x)
y21 <- cos(2 * x)
y12 <- sin(x)
y22 <- sin(2 * x)
df <- data.frame(id = factor(rep(1:2, each = length(x))),
x = rep(x, times = 2),
y1 = c(y11, y21),
y2 = c(y12, y22))
mfdobj <- get_mfd_df(dt = df,
domain = c(1, 10),
arg = "x",
id = "id",
variables = c("y1", "y2"),
lambda = 1e-5)