bada: Barycentric Discriminant Analysis

In bbuchsbaum/neuropls: neuroca: multiblock component analysis for neuroimaging data

Barycentric Discriminant Analysis

Description

A component technique that maximizes the between group variance over a set of variables.

Usage

bada(
  Y,
  X,
  S = rep(1, nrow(X)),
  ncomp = length(levels(as.factor(Y))) - 1,
  preproc = center(),
  A = NULL,
  M = NULL,
  ...
)

Arguments

Y	dependent factor variable. The categories of the observations.
X	the data matrix with nrow equal to length of Y
S	an integer vector or factor of subject ids which is the same length as Y.
ncomp	number of components to estimate
preproc	pre-processing function, defaults to center
A	the column constraints
M	the row constraints
...	arguments to pass through

Details

The S argument can be used to model multi-level structure. If S is included, then pre-processing is applied separately to each unique value of S. This has the effect, for example when preproc = center of removing subject-specific means before computing the barycetners.

References

Abdi, H., Williams, L. J., & Bera, M. (2017). Barycentric discriminant analysis. Encyclopedia of Social Network Analysis and Mining, 1-20.

Examples

X <- matrix(rnorm(100*1000), 100, 1000)
Y <- factor(rep(letters[1:4], length.out=100))
S <- factor(rep(1:10, each=10))

bres <- bada(Y, X, S, ncomp=3)

project(bres, X[S==1,])

## no strata
bres <- bada(Y, X, ncomp=3)



xbar <- matrix(rnorm(4*1000), 4, 1000)
Y <- factor(rep(letters[1:4], length.out=100))
X <- do.call(rbind, lapply(as.character(Y), function(y) { switch(y,
                               "a" = xbar[1,] + rnorm(1000)*5,
                               "b" = xbar[2,] + rnorm(1000)*5,
                               "c" = xbar[3,] + rnorm(1000)*5,
                               "d" = xbar[4,] + rnorm(1000)*5)
                               }
                               ))
                               
bres2 <- bada(Y, X, S, ncomp=3)

bbuchsbaum/neuropls documentation built on April 17, 2022, 8:46 a.m.