Applies accelerated proximal gradient algorithm to the optimal scoring formulation of sparse discriminant analysis proposed by Clemmensen et al. 2011.
SDAAP(Xt, ...)# S3 method for default
SDAAP(
Xt,
Yt,
Om,
gam,
lam,
q,
PGsteps,
PGtol,
maxits,
tol,
selector = rep(1, dim(Xt)[2]),
initTheta,
bt = FALSE,
L,
eta,
rankRed = FALSE,
...
)
SDAAP returns an object of class "SDAAP" including a list
with the following named components:
callThe matched call.
Bp by q matrix of discriminant vectors.
QK by q matrix of scoring vectors.
subitsTotal number of iterations in proximal gradient subroutine.
totalitsNumber coordinate descent iterations for all discriminant vectors
NULL
n by p data matrix, (not a data frame, but a matrix)
n by K matrix of indicator variables (Yij = 1 if i in class j). This will later be changed to handle factor variables as well. Each observation belongs in a single class, so for a given row/observation, only one element is 1 and the rest is 0.
p by p parameter matrix Omega in generalized elastic net penalty.
Regularization parameter for elastic net penalty.
Regularization parameter for l1 penalty, must be greater than zero.
Desired number of discriminant vectors.
Maximum number if inner proximal gradient algorithm for finding beta.
Stopping tolerance for inner APG method.
Number of iterations to run
Stopping tolerance for proximal gradient algorithm.
Vector to choose which parameters in the discriminant vector will be used to calculate the regularization terms. The size of the vector must be *p* the number of predictors. The default value is a vector of all ones. This is currently only used for ordinal classification.
Option to set the initial theta vector, by default it is a vector of all ones for the first theta.
Boolean to indicate whether backtracking should be used, default false.
Initial estimate for Lipshitz constant used for backtracking.
Scalar for Lipshitz constant.
Boolean indicating whether Om is in factorized form, such that R^t*R = mO
SDAAPcv, SDAP and SDAD