irlsl1reg {PEIP}

L1 least squares with sparsity

Description

Solves the system Gm=d using sparsity regularization on Lm. Solves the L1 regularized least squares problem: min norm(G*m-d,2)^2+alpha*norm(L*m,1)

Usage

irlsl1reg(G, d, L, alpha, maxiter = 100, tolx = 1e-04, tolr = 1e-06)

Arguments

G

design matrix

d

right hand side

L

regularization matrix

alpha

regularization parameter

maxiter

Maximum number of IRLS iterations

tolx

Tolerance on successive iterates

tolr

Tolerance below which we consider an element of L*m to be effectively zero

Value

m

model vector

Author(s)

Jonathan M. Lees<[e-mail]>

References

Aster, R.C., C.H. Thurber, and B. Borchers, Parameter Estimation and Inverse Problems, Elsevier Academic Press, Amsterdam, 2005.

Examples

n = 20
G = shawG(n,n)

spike = rep(0,n)
spike[10] = 1

spiken = G %*%  spike

wts = rep(1, n)
delta = 1e-03
set.seed(2015)
dspiken = spiken + 6e-6 *rnorm(length(spiken))
L1 = get_l_rough(n,1);
alpha = 0.001

k = irlsl1reg(G, dspiken, L1, alpha, maxiter = 100, tolx = 1e-04, tolr = 1e-06)


plotconst(k,-pi/2,pi/2, ylim=c(-.2,  0.5), xlab="theta", ylab="Intensity" );