EPSODE: Perform the generic Path Algorithm for a LVM

In bozenne/lava.penalty: Penalized Latent Variable Models

Description

Perform the generic Path Algorithm for a LVM

Usage

EPSODE(start, objective, gradient, hessian, V, lambda2, index.penalty2,
  equivariance, constrain.variance, index.variance, control)

Arguments

start	the starting value
objective	likelihood given by lava. Used to adjust the step parameter when using backtracking
gradient	first derivative of the likelihood given by lava.
hessian	second derivative of the likelihood given by lava. Only used to estimate the step parameter of the algorithm when step = NULL
V	matrix that left multiply beta to define the penalization (identity corresponds to a standard lasso penalty)
lambda2	ridge penalization parameter
index.penalty2	parameters to which ridge penalization is applied
equivariance	should the lambda parameter be multiplied with the first variance parameter?
constrain.variance	should the variance parameters be log transformed?
index.variance	the position of the variance parameters in start
control	settings for the EPSODE algorithm. See lava.options.

Details

Does not work for an unknown variance matrix since the log-likelihood of the regression and variance parameter is not jointly convex (according to ??)

References

Zhou 2014 - A generic Path Algorithm for Regularized Statistical Estimation

bozenne/lava.penalty documentation built on May 13, 2019, 1:41 a.m.