hero (version 0.6)

prepare.matrix: Prepare data matrix for sandwich smooth

Description

prepare.matrix prepares a data matrix for the sandwich smooth. The dimensionality of data and the length of x must match. Specifically, length(dim(data)) must equal length(x). The dimensionality of data and the length of splines must match. Specifically, length(dim(data)) must equal length(splines).

Usage

# S3 method for matrix
prepare(
  data,
  x,
  splines,
  m = 2,
  sparse = TRUE,
  spdiffpen = TRUE,
  digits = 1,
  sts = FALSE,
  ...
)

Value

A prepared_matrix object.

Arguments

data

A data matrix.

x

A list of values at which to evaluate the basis functions. See Examples and Details.

splines

A list of spline objects (hero_bspline and hero_radspline). See Examples and Details.

m

A positive integer indicating order of the difference penalty.

sparse

A logical value indicating if the result should be a sparse version of the Matrix-class.

spdiffpen

A logical value indicating whether spdiffpen should be used to compute the difference penalty. The default is FALSE.

digits

The number of digits to use when applying round to the distances.

sts

A logical value indicating whether data is a spatial time series, in which each row of data corresponds to a distinct spatial location and each column corresponds to a distinct time.

...

Not currently implemented.

Author

Joshua French. Based off code by Luo Xiao (see References).

Details

For a typical sandwich smooth (sts = FALSE), for two-dimensional data, data[i, j] is assumed to be observed at position x[[1]][i], x[[2]][j]. If the data are a spatial time series, then the first dimension is assumed to refer to space, and the second dimension to time. In that case, data[i, j] is assumed to be observed at location x[[1]][i, ] and time x[[2]][j].

If sts = TRUE, then x[[1]] should be a matrix of spatial coordinates, with each row corresponding to a location, and x[[2]] should be a vector with the observation times.

If x is not supplied, then default.evalargs is used to create it automatically. This is only valid when sts = FALSE.

If splines is not supplied, then a B-spline basis is automatically created for each dimension using default.splines. This is only valid when sts = FALSE.

References

Xiao, L. , Li, Y. and Ruppert, D. (2013), Fast bivariate P-splines: the sandwich smoother. J. R. Stat. Soc. B, 75: 577-599. <doi:10.1111/rssb.12007>

See Also

bspline, radspline, diffpen, spdiffpen, default.evalargs, default.splines

Examples

# prepare Lu et al. (2012) noisy f1 data
data(ludata)
obj = prepare(lunoisyf1, x = list(x, z))
h = hero(obj)

# precompute some stuff
splines = default.splines(list(x, z))
l = assemble(splines, x = list(x, z))
obj2 = prepare(lunoisyf1, x = list(x, z),
               splines = splines, assembled = l)
h2 = hero(obj2)
all.equal(h, h2)