Robust Discriminant Analysis

DESCRIPTION:

Fit a robust linear or quadratic discriminant function to a set of feature data.

USAGE:

discRob(formula, data=sys.parent(), family=Classical("homoscedastic"),
        frequencies, na.action=na.exclude, subset,
        prior= c("proportional", "uniform", "none"),
        method=c("svd", "choleski"), singular.tol=sqrt(.Machine$double.eps),
        estim="auto", cov.control=covRob.control(estim=estim), ...)

REQUIRED ARGUMENTS:

formula
a formula object, specifying the group variable and feature variables, with the group variable on the left of a `~' operator, and the feature variables, separated by + operators, on the right. If data is given, all names used in the formula should be defined as variables in the data frame.

OPTIONAL ARGUMENTS:

data
a data frame in which to interpret the variables named in formula. Default is the calling frame. This is commonly refered to as the training data for the discriminant function.
family
a family.discrim object. Currently, only the Classical constructor with its argument cov.structure equal to "homoscedastic" or "heteroscedastic" is implemented.
frequencies
a vector of observation frequencies.
na.action
a function to filter missing data. This is applied to the model frame after any subset argument has been used. The default na.exclude is to delete the observation if any missing values are found. A possible alternative is na.fail, which generates an error if any missing values are found.
subset
expression specifying a row subset of the data to be used in the fit (training data). This can be a logical vector, or a numeric vector indicating which observation numbers are to be included, or a character vector with the row names to be included. All observations are included by default.
prior
a character string or numerical vector specifying the prior knowledge of the mixing proportions of each group. The acceptable strings are as follows: "proportional", group of proportions are the number of observations from each group divided by the total number of observations; "uniform", group proportions are one over the number of groups; "none", exclude the mixing proportion from the discriminant function. If prior is a numerical vector, it must have a length equal to the number of groups and its elements must be positive and sum to one.
method
numerical method to be used to decompose the feature matrix of covariances. The choices are "svd", singular value decomposition (the default) and "choleski", Choleski decomposition.
singular.tol
tolerance for determining existing linear dependencies among the feature vectors.
estim
the robust estimator used by covRob. The choices are: "mcd" for the Fast MCD algorithm of Rousseeuw and Van Driessen, "donostah" for the Donoho-Stahel projection based estimator, "M" for the constrained M estimator provided by Rocke, "pairwiseQC" for the quadrant correlation based pairwise estimator, and "pairwiseGK" for the Gnanadesikan-Kettenring pairwise estimator. The default "auto" selects from "donostah", "mcd", and "pairwiseQC" with the goal of producing a good estimate in a resonable amount of time.
cov.control
a list of control parameters to be used in the numerical algorithms. See covRob.control for the possible control parameters and their default settings.

VALUE:

an object of class discRob that contains the robust discriminant function

SEE ALSO:

, .

EXAMPLES:

discRob(Group ~ ., data = hemo.cont, family = Classical("heter"))