prox_bcd: R Wrapper to C Function for GGB-Local Penalty Proximal...

In jacobbien/ggb: Graph-Guided Banding of the Covariance Matrix

Description

Returns sumvv which can be thought of as a (p choose 2)-by-nlam matrix, each column being the lower triangle of sum_jb V_jb.

Usage

prox_bcd(R, D, M, lambda = NULL, nlam = 20, flmin = 0.01, maxiter = 100,
  tol = 1e-04, verbose = 1)

Arguments

R	a symmetric, p by p matrix
D	matrix of graph distances
M	p-vector of maximal bandwidths
lambda	a vector of positive tuning parameters. If NULL, uses nlam, flmin to produce a grid (formation of lambda grid in this case is done within C)
nlam	number of lambda values (ignored if lambda is nonnull)
flmin	ratio of smallest to largest lambda value (ignored if lambda is nonnull)
maxiter	maximum number of passes through coordinates
tol	convergence threshold
verbose	level of verbosity in printed output (2, 1, 0)

Details

argmin 0.5 * || R - sum_jb V_jb ||_F^2 + lam * sum_jb w_jb || V_jb ||_F where V_jb is 0 off of g_jb.

jacobbien/ggb documentation built on May 18, 2019, 8:01 a.m.