tfpca: Temporal Functional Principal Component

In guiludwig/stdf: Spatio-Temporal Data Fusion Model

Description

This function is used on step II of the STDF algorithm. It does not need to be used directly, since it's mostly an wrapper to fda package function calls. It behaves similarly to the pca.fd function, but rescales the eigenfunctions/eigenvalues in a way that's needed by the stdf function. Notice tfpca do not handle NA's in the data values. It might be useful do to some kind of nearest neightbor imputation if possible. The fda function data2fd does a least-squares fit of the basis function, which can be helpful in preprocessing the data. The tuning parameter zeta must be passed manually.

Usage

tfpca(MatY, L, t.fit, zeta = 0, tbas = 20)

Arguments

MatY	The data organized in a matrix, with each time series in a column. Missing values are currently not handled properly.
L	The number of functional principal components to be returned.
t.fit	Values of the time-domain at which a value is fitted.
zeta	Smoothness parameter, must be provided. Defaults to 0 (no smoothing).
tbas	Number of spline basis functions used in the smooth.

Value

List of three items

values	numeric vector of L Eigenvalue estimates
vectors	matrix of L columns with Eigenfunction estimates observed at t.fit
harmfd	Internal use. Prediction of the Eigenfunction at new values of t, say under vector t.pred, can be obtained with the next three parameters. The syntax is fda::eval.fd(t.pred, harmfd)*t(matrix(sqrt(nObs/etan), L, length(t.pred)))
nObs	Internal use.
etan	Internal use.

References

Ramsay, J. O. (2006) Functional Data Analysis. New York: Springer.

See Also

pca.fd, stdf

Examples

# This example uses fda's CanadianWeather dataset:
library(fda)
test <- tfpca(CanadianWeather$dailyAv[,,"Temperature.C"], 2, 1:365, 0, 20)
test2 <- tfpca(CanadianWeather$dailyAv[,,"Temperature.C"], 2, 1:365, 1e5, 20)
layout(matrix(1:4, byrow = TRUE, ncol = 2))
plot(test$vectors[,1], type="l", main="First Principal Component")
plot(test$vectors[,2], type="l", main="Second Principal Component")
plot(test2$vectors[,1], type="l", main="Smooth First Principal Component")
plot(test2$vectors[,2], type="l", main="Smooth Second Principal Component")

guiludwig/stdf documentation built on May 17, 2019, 9:27 a.m.