PFC: Principal Fitted Components Regression

In abnormally-distributed/cvreg: Cross Validation and Robust Estimation Utilities

Description

Principal fitted components (PFC) regression was developed to address two flaws with principal components regression (Cook & Forzani, 2008). First, in PCR the components are computed from the predictors alone and do not make apparent use of the response – they are simply the principal components of the design matrix. Second, the principal components are not invariant or equivariant under full rank linear transformation of the predictors.

Usage

PFC(
  formula,
  data,
  slices = 3,
  rank = "all",
  ytype = c("numeric", "categorical"),
  sfun = c("gp", "ns", "bs", "cr", "poly")
)

Arguments

formula	a model formula
data	a data frame
slices	the number of slices to use. technically in this function the slices are produced by a spline function with df = slices. the default is 3.
rank	the desired number of sufficient predictors to return
ytype	either numeric or categorical
sfun	for numeric variables, the basis function to use. must be one of "gp" (the default), "ns", "bs", "cr", or "poly", which respectively correspond to gaussian process, natural splines, basis splines, cubic regression splines, and orthogonal polynomials.

Value

an sdr object

References

Cook, R.D., Forzani, L. (2008) Principal Fitted Components for Dimension Reduction in Regression. Statist. Sci. 23(4), 485-501. doi:10.1214/08-STS275.

abnormally-distributed/cvreg documentation built on May 3, 2020, 3:45 p.m.