update_Sigma_pgd: Accelerated projected gradient descent update function for...

In koekvall/lvmmr: Mixed-type multivariate response regression

Description

Accelerated projected gradient descent update function for Sigma

Usage

update_Sigma_pgd(
  R,
  D2,
  psi,
  Sigma.init,
  M,
  epsilon = 0,
  tol.dykstra = 1e-12,
  tol.ipiano = 1e-10,
  max.iter.dykstra = 1000,
  max.iter.ipiano = 1000,
  quiet = TRUE
)

Arguments

R	Matrix (n x r) of working residuals.
D2	Matrix (n x p) second derivatives of cumulant functions evaluated at the expansion points.
psi	Vector (r x 1) of conditional variance parameters.
Sigma.init	Matrix (r x r ) with starting value for Sigma.
M	Matrix (r x r) with restrictions for elements o Sigma (NA = no restriction).
epsilon	Scalar with lower bound on eigenvalues of Sigma.
tol.dykstra	Scalar tolerance parameter for projection algorithm,
tol.ipiano	Scalar tolerance for descent algorithm.
max.iter.dykstra	Integer maximum number of iterations for projection algorithm.
max.iter.ipiano	Integer maximum number of iterations for descent algorithm.
quiet	Logical indicating whether to print information while running.

Value

Matrix (r x r) with update of Sigma

koekvall/lvmmr documentation built on Dec. 13, 2021, 2:35 p.m.