daCoDa: Linear and quadratic discriminant analysis for compositional...

View source: R/daCoDa.R

daCoDaR Documentation

Linear and quadratic discriminant analysis for compositional data.

Description

Linear and quadratic discriminant analysis for compositional data using either robust or classical estimation.

Usage

daCoDa(x, grp, coda = TRUE, method = "classical", rule = "linear", ...)

Arguments

x

a matrix or data frame containing the explanatory variables

grp

grouping variable: a factor specifying the class for each observation.

coda

TRUE, when the underlying data are compositions.

method

“classical” or “robust”

rule

a character, either “linear” (the default) or “quadratic”.

...

additional arguments for the functions passed through

Details

Compositional data are expressed in orthonormal (ilr) coordinates (if coda==TRUE). For linear discriminant analysis the functions LdaClassic (classical) and Linda (robust) from the package rrcov are used. Similarly, quadratic discriminant analysis uses the functions QdaClassic and QdaCov (robust) from the same package.

The classical linear and quadratic discriminant rules are invariant to ilr coordinates and clr coefficients. The robust rules are invariant to ilr transformations if affine equivariant robust estimators of location and covariance are taken.

Value

An S4 object of class LdaClassic, Linda, QdaClassic or QdaCov. See package rrcov for details.

Author(s)

Jutta Gamper

References

Filzmoser, P., Hron, K., Templ, M. (2012) Discriminant analysis for compositional data and robust parameter estimation. Computational Statistics, 27(4), 585-604.

See Also

LdaClassic, Linda, QdaClassic, QdaCov

Examples

## toy data (non-compositional)
require(MASS)
x1 <- mvrnorm(20,c(0,0,0),diag(3))
x2 <- mvrnorm(30,c(3,0,0),diag(3))
x3 <- mvrnorm(40,c(0,3,0),diag(3))
X <- rbind(x1,x2,x3)
grp=c(rep(1,20),rep(2,30),rep(3,40))

clas1 <- daCoDa(X, grp, coda=FALSE, method = "classical", rule="linear")
summary(clas1)
## predict runs only with newest verison of rrcov
## Not run: 
predict(clas1)

## End(Not run)
# specify different prior probabilities
clas2 <- daCoDa(X, grp, coda=FALSE, prior=c(1/3, 1/3, 1/3))
summary(clas2)

## compositional data
data(coffee)
x <- coffee[coffee$sort!="robusta",2:7]
group <- droplevels(coffee$sort[coffee$sort!="robusta"])
cof.cla <- daCoDa(x, group, method="classical", rule="quadratic")
cof.rob <- daCoDa(x, group, method="robust", rule="quadratic")
## predict runs only with newest verison of rrcov
## Not run: 
predict(cof.cla)@ct
predict(cof.rob)@ct

## End(Not run)

robCompositions documentation built on Aug. 25, 2023, 5:13 p.m.