Type: Package
Title: Flexible Regularized Estimating Equations
Version: 1.0.2
Date: 2024-05-22
Description: Unified regularized estimating equation solver. Currently the package includes one solver with the l1 penalty only. More solvers and penalties are under development. Reference: Yi Yang, Yuwen Gu, Yue Zhao, Jun Fan (2021) <doi:10.48550/arXiv.2110.11074>.
License: GPL-3
Imports: Rcpp (≥ 1.0.7)
LinkingTo: Rcpp, RcppArmadillo
Encoding: UTF-8
RoxygenNote: 7.3.1
Suggests: testthat (≥ 3.0.0)
Config/testthat/edition: 3
NeedsCompilation: yes
Packaged: 2024-05-24 02:30:24 UTC; YLIAN
Author: Yi Lian [aut, cre], Yi Yang [aut, cph], Yuwen Gu [aut], Jun Fan [aut], Yue Zhao [aut], Robert W. Platt [aut]
Maintainer: Yi Lian <yi.lian@mail.mcgill.ca>
Repository: CRAN
Date/Publication: 2024-05-24 04:00:02 UTC

free: Flexible Regularized Estimating Equations

Description

Unified regularized estimating equation solver. Currently the package includes one solver with the l1 penalty only. More solvers and penalties are under development. Reference: Yi Yang, Yuwen Gu, Yue Zhao, Jun Fan (2021) doi:10.48550/arXiv.2110.11074.

Author(s)

Maintainer: Yi Lian yi.lian@mail.mcgill.ca

Authors:

Main solver of free

Description

Main solver of free

Usage

free_lasso(
  p,
  lambda,
  est_func,
  par_init,
  alpha,
  tau,
  maxit = 1000L,
  tol_ee = 1e-06,
  tol_par = 1e-06,
  verbose = FALSE
)

Arguments

p

The dimension of the dataset

lambda

Lasso regularization coefficient

est_func

R function, the estimating function specified by the user

par_init

Optional, initial value for parameter update

alpha

Tuning parameter

tau

Tuning parameter

maxit

Maximum iterations

tol_ee

Convergence criterion based on the update of the estimating function

tol_par

Convergence criterion based on the update of the parameter

verbose

logical, print updates

Value

A list containing the regularized estimating equation estimates and the number of iterations it takes to converge.

Examples

# Standardize data
dat <- scale(mtcars)
x <- as.matrix(dat[, -1])
y <- as.vector(dat[, 1])
n <- nrow(x)
p <- ncol(x)

# Specify estimating function
ufunc <- function(b) {
  1/n * crossprod(x, (x %*% b - y) )
}

# Set hyperparameters
tau <- 0.6
alpha <- 0.5

# Set regularization coefficient
lambda1 <- 0
free_R1 <- free_lasso(p = p,
                      lambda = lambda1,
                      est_func = ufunc,
                      par_init = rep(0, p),
                      alpha = alpha,
                      tau = tau,
                      maxit = 10000L,
                      tol_ee = 1e-20,
                      tol_par = 1e-10,
                      verbose = FALSE)
free_R1$coefficients

# Compare with lm() - very close
lm(y~x-1)$coefficients

# Set regularization coefficient
lambda2 <- 0.7
free_R2 <- free_lasso(p = p,
                      lambda = lambda2,
                      est_func = ufunc,
                      par_init = rep(0, p),
                      alpha = alpha,
                      tau = tau,
                      maxit = 10000L,
                      tol_ee = 1e-20,
                      tol_par = 1e-10,
                      verbose = FALSE)
free_R2$coefficients