succotash_unif_fixed: A fixed point iteration in the mixture of uniforms EM.

In dcgerard/succotashr: Surrogate and Confounder Correction Occuring Together with Adaptive Shrinkage

Description

This is a fixed-point iteration for the SUCCOTASH EM algorithm. This updates the estimate of the prior and the estimate of the hidden covariates.

Usage

succotash_unif_fixed(pi_Z, lambda, alpha, Y, a_seq, b_seq, sig_diag,
  print_ziter = FALSE, newt_itermax = 4, tol = 10^-4, var_scale = TRUE)

Arguments

pi_Z	A vector. The first M values are the current values of π. The last k values are the current values of Z.
lambda	A vector. This is a length M vector with the regularization parameters for the mixing proportions.
alpha	A matrix. This is of dimension p by k and are the coefficients to the confounding variables.
Y	A matrix of dimension p by 1. These are the observed regression coefficients of the observed variables.
a_seq	A vector of negative numerics containing the left endpoints of the mixing uniforms.
b_seq	A vector of positiv numerics containing the right endpoints of the mixing uniforms.
sig_diag	A vector of length p containing the variances of the observations.
print_ziter	A logical. Should we we print each iteration of the Z optimization?
newt_itermax	A positive integer. The maximum number of Newton steps to perform in updating Z.
tol	A positive numeric. The stopping criterion for Newton's method in updating Z.
var_scale	A logical. Should we update the scaling on the variances (TRUE) or not (FALSE).

Value

pi_new A vector of length M. The update for the mixing components.

Z_new A vector of length k. The update for the confounder covariates.

See Also

uniform_succ_given_alpha succotash_llike_unif

dcgerard/succotashr documentation built on May 15, 2019, 1:25 a.m.