1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133
|
\name{covOGK}
\alias{covOGK}
\alias{covGK}
\alias{s_mad}
\alias{s_IQR}
\alias{hard.rejection}
%
\title{Orthogonalized Gnanadesikan-Kettenring (OGK) Covariance Matrix Estimation}
\description{
Computes the orthogonalized pairwise covariance matrix estimate described in
in Maronna and Zamar (2002). The pairwise proposal goes back to
Gnanadesikan and Kettenring (1972).
}
\usage{
covOGK(X, n.iter = 2, sigmamu, rcov = covGK, weight.fn = hard.rejection,
keep.data = FALSE, \dots)
covGK (x, y, scalefn = scaleTau2, \dots)
s_mad(x, mu.too = FALSE, na.rm = FALSE)
s_IQR(x, mu.too = FALSE, na.rm = FALSE)
}
\arguments{
\item{X}{data in something that can be coerced into a numeric matrix.}
\item{n.iter}{number of orthogonalization iterations. Usually 1 or 2;
values greater than 2 are unlikely to have any significant effect on
the estimate (other than increasing the computing time).}
\item{sigmamu, scalefn}{a function that computes univariate robust
location and scale estimates. By default it should return a single
numeric value containing the robust scale (standard deviation)
estimate. When \code{mu.too} is true, \code{sigmamu()} should
return a numeric vector of length 2 containing robust location and
scale estimates. See \code{\link{scaleTau2}}, \code{\link{s_Qn}},
\code{\link{s_Sn}}, \code{s_mad} or \code{s_IQR} for examples to be
used as \code{sigmamu} argument.}
\item{rcov}{function that computes a robust covariance estimate
between two vectors. The default, Gnanadesikan-Kettenring's
\code{covGK}, is simply \eqn{(s^2(X+Y) - s^2(X-Y))/4} where
\eqn{s()} is the scale estimate \code{sigmamu()}.}
\item{weight.fn}{a function of the robust distances and the number of
variables \eqn{p} to compute the weights used in the reweighting step.}
\item{keep.data}{logical indicating if the (untransformed) data matrix
\code{X} should be kept as part of the result.}
\item{\dots}{additional arguments; for \code{covOGK} to be passed to
\code{sigmamu()} and \code{weight.fn()}; for \code{covGK} passed to \code{scalefn}.}
%%% covGK():
\item{x,y}{numeric vectors of the same length, the covariance of which
is sought in \code{covGK} (or the scale, in \code{s_mad} or
\code{s_IQR}).}
%%% s_mad(), s_IQR():
\item{mu.too}{logical indicating if both location and scale should be
returned or just the scale (when \code{mu.too=FALSE} as by default).}
\item{na.rm}{if \code{TRUE} then \code{\link{NA}} values are stripped
from \code{x} before computation takes place.}
}
\details{
Typical default values for the \emph{function} arguments
\code{sigmamu}, \code{rcov}, and \code{weight.fn}, are
available as well, see the \emph{Examples} below,
\bold{but} their names and calling sequences are
still subject to discussion and may be changed in the future.
The current default, \code{weight.fn = hard.rejection} corresponds to
the proposition in the litterature, but Martin Maechler strongly
believes that the hard threshold currently in use is too arbitrary,
and further that \emph{soft} thresholding should be used instead, anyway.
}
\value{
\code{covOGK()} currently returns a list with components
\item{center}{robust location: numeric vector of length \eqn{p}.}
\item{cov}{robust covariance matrix estimate: \eqn{p\times p}{p x p}
matrix.}
\item{wcenter, wcov}{re-\bold{w}eighted versions of \code{center} and
\code{cov}.}
\item{weights}{the robustness weights used.}
\item{distances}{the mahalanobis distances computed using
\code{center} and \code{cov}.}
\dots\dots
\cr
\bold{but note that this might be radically changed to returning an
S4 classed object!}
\code{covGK()} is a trivial 1-line function returning the covariance
estimate
\deqn{\hat c(x,y) = \left(\hat \sigma(x+y)^2 - \hat \sigma(x-y)^2 \right)/4,%
}{ c^(x,y) = [s^(x+y)^2 - s^(x-y)^2]/4,}%
where \eqn{\hat \sigma(u)}{s^(u)} is the scale estimate of \eqn{u}
specified by \code{scalefn}.
\code{s_mad()}, and \code{s_IQR()} return the
scale estimates \code{\link[stats]{mad}} or \code{\link[stats]{IQR}}
respectively, where the \code{s_*} functions return a length-2 vector
(mu, sig) when \code{mu.too = TRUE}, see also \code{\link{scaleTau2}}.
}
\references{
Maronna, R.A. and Zamar, R.H. (2002)
Robust estimates of location and dispersion of high-dimensional datasets;
\emph{Technometrics} \bold{44}(4), 307--317.
Gnanadesikan, R. and John R. Kettenring (1972)
Robust estimates, residuals, and outlier detection with multiresponse data.
\emph{Biometrics} \bold{28}, 81--124.
}
\author{Kjell Konis \email{konis@stats.ox.ac.uk}, with modifications by
Martin Maechler.}
\seealso{\code{\link{scaleTau2}},
\code{\link{covMcd}}, \code{\link[MASS]{cov.rob}}.
}
\examples{
data(hbk)
hbk.x <- data.matrix(hbk[, 1:3])
cO1 <- covOGK(hbk.x, sigmamu = scaleTau2)
cO2 <- covOGK(hbk.x, sigmamu = s_Qn)
cO3 <- covOGK(hbk.x, sigmamu = s_Sn)
cO4 <- covOGK(hbk.x, sigmamu = s_mad)
cO5 <- covOGK(hbk.x, sigmamu = s_IQR)
%% FIXME: Add time comparison, here or in "vignette", "demo", "...
data(toxicity)
cO1tox <- covOGK(toxicity, sigmamu = scaleTau2)
cO2tox <- covOGK(toxicity, sigmamu = s_Qn)
## nice formatting of correlation matrices:
as.dist(round(cov2cor(cO1tox$cov), 2))
as.dist(round(cov2cor(cO2tox$cov), 2))
## "graphical"
symnum(cov2cor(cO1tox$cov))
symnum(cov2cor(cO2tox$cov), legend=FALSE)
}
\keyword{robust}
\keyword{multivariate}
|