Help for package LassoHiDFastGibbs

Type:

Package

Title:

Fast High-Dimensional Gibbs Samplers for Bayesian Lasso Regression

Version:

0.1.4

Description:

Provides fast and scalable Gibbs sampling algorithms for Bayesian Lasso regression model in high-dimensional settings. The package implements efficient partially collapsed and nested Gibbs samplers for Bayesian Lasso, with a focus on computational efficiency when the number of predictors is large relative to the sample size. Methods are described at Davoudabadi and Ormerod (2026) https://github.com/MJDavoudabadi/LassoHiDFastGibbs.

License:

GPL-3

Encoding:

UTF-8

SystemRequirements:

C++17

Imports:

Rcpp

LinkingTo:

Rcpp, RcppArmadillo, RcppEigen, RcppNumerical, RcppClock

NeedsCompilation:

yes

RoxygenNote:

7.3.2

Suggests:

posterior

URL:

https://github.com/MJDavoudabadi/LassoHiDFastGibbs

BugReports:

https://github.com/MJDavoudabadi/LassoHiDFastGibbs/issues

Packaged:

2026-01-26 01:33:54 UTC; mjava

Author:

John Ormerod

[aut], Mohammad Javad Davoudabadi [aut, cre, cph], Garth Tarr

[aut], Samuel Mueller

[aut], Jonathon Tidswell [ctb] (Contributed code to src/lasso_distribution.cpp (originally from BayesianLasso package))

Maintainer:

Mohammad Javad Davoudabadi <mohammad.davoudabadi@qut.edu.au>

Repository:

CRAN

Date/Publication:

2026-01-29 21:50:14 UTC

Fast High-Dimensional Gibbs Samplers for Bayesian Lasso Regression

Description

Author(s)

Maintainer: Mohammad Javad Davoudabadi mohammad.davoudabadi@qut.edu.au [copyright holder]

Authors:

John Ormerod john.ormerod@sydney.edu.au (ORCID)
Garth Tarr garth.tarr@gmail.com (ORCID)
Samuel Mueller samuel.mueller@sydney.edu.au (ORCID)

Other contributors:

Jonathon Tidswell (Contributed code to src/lasso_distribution.cpp (originally from BayesianLasso package)) [contributor]

Bayesian lasso Gibbs sampler: 2-block (beta–lambda2) variant

Description

Implements a two-block Gibbs sampler for the Bayesian lasso regression model in which the regression coefficients are updated jointly with the global shrinkage parameter \lambda^2 in one block, while the noise variance and local shrinkage parameters are updated conditionally in separate steps.

Usage

blasso_gibbs_2block_bl(
  vy,
  mX,
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  va_init = NULL,
  verbose = max(1L, floor(nsamples/5)),
  lower = 1e-12,
  upper = 5000
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

a, b

Hyperparameters for the inverse-gamma prior on sigma^2.

u, v

Hyperparameters for the prior on lambda^2.

nsamples

Integer number of MCMC iterations.

lambda_init

Initial value for lambda.

sigma2_init

Initial value for sigma^2.

va_init

Optional initial values for local shrinkage parameters (length p).

verbose

Print progress every verbose iterations (0 = silent).

lower, upper

Bounds used by the slice sampler for lambda^2.

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 30; p <- 6
X <- matrix(rnorm(n * p), n, p)
y <- rnorm(n)
out <- blasso_gibbs_2block_bl(
  vy = y, mX = X,
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  verbose = 0
)
str(out)

Bayesian lasso Gibbs sampler: 2-block (beta–sigma2) variant

Description

Implements a two-block Gibbs sampler for the Bayesian lasso regression model in which the regression coefficients are updated jointly with the noise variance \sigma^2 in one block, while the global shrinkage parameter and local shrinkage parameters are updated conditionally in separate steps.

Usage

blasso_gibbs_2block_bs(
  vy,
  mX,
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  verbose = max(1L, floor(nsamples/5))
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

a, b

Hyperparameters for the inverse-gamma prior on sigma^2.

u, v

Hyperparameters for the prior on lambda^2.

nsamples

Integer number of MCMC iterations.

lambda_init

Initial value for lambda.

sigma2_init

Initial value for sigma^2.

verbose

Print progress every verbose iterations (0 = silent).

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 30; p <- 6
X <- matrix(rnorm(n * p), n, p)
y <- rnorm(n)
out <- blasso_gibbs_2block_bs(
  vy = y, mX = X,
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  verbose = 0
)
str(out)

Bayesian lasso PCG sampler: lambda2 collapsed over local scales

Description

Lasso-specific Partially-collapsed Gibbs (PCG) variant with the local scales (va) collapsed in the \lambda^2 update.

Usage

blasso_pcg_lambda2_va(
  vy,
  mX,
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  verbose = max(1L, floor(as.integer(nsamples)/10))
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Number of MCMC iterations.

lambda_init

Initial value for \lambda.

sigma2_init

Initial value for \sigma^2.

verbose

Print progress every verbose iterations (0 = silent).

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 40; p <- 6
X <- matrix(rnorm(n * p), n, p)
beta <- c(1.2, 2, -1, 0.5, 0.75, 2.5)
y <- X %*% beta + rnorm(n)

out <- blasso_pcg_lambda2_va(
  vy = y, mX = X,
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  verbose = 0
)

summary(out$vlambda2)

Bayesian lasso PCG sampler: sigma2 collapsed over local scales

Description

Lasso-specific Partially-collapsed Gibbs (PCG) variant with the local scales (va) collapsed in the \sigma^2 update.

Usage

blasso_pcg_sigma2_va(
  vy,
  mX,
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  va_init = NULL,
  verbose = max(1L, floor(as.integer(nsamples)/10)),
  lower = 1e-12,
  upper = 5000
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Number of MCMC iterations.

lambda_init

Initial value for \lambda.

sigma2_init

Initial value for \sigma^2.

va_init

Optional initial local-scale vector (length p). If NULL, defaults to rep(1, ncol(mX)).

verbose

Print progress every verbose iterations (0 = silent).

lower, upper

Bounds used by the slice sampler.

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 40; p <- 6
X <- matrix(rnorm(n * p), n, p)
beta <- c(1.2, 2, -1, 0.5, 0.75, 2.5)
y <- X %*% beta + rnorm(n)

out <- blasso_pcg_sigma2_va(
  vy = y, mX = X,
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  va_init = rep(1, p),
  verbose = 0
)

summary(out$vsigma2)

Normalize Response and Covariates

Description

This function centers and (optionally) scales the response vector and each column of the design matrix using the population variance. It is used to prepare data for Bayesian Lasso regression.

Usage

normalize(y, X, scale = TRUE)

Arguments

y

A numeric response vector.

X

A numeric matrix or data frame of covariates (design matrix).

scale

Logical; if TRUE, variables are scaled to have unit population variance (default is TRUE).

Value

A list with the following elements:

vy: Normalized response vector.
mX: Normalized design matrix.
mu.y: Mean of the response vector.
sigma2.y: Population variance of the response vector.
mu.x: Vector of column means of X.
sigma2.x: Vector of population variances for columns of X.

Examples

set.seed(1)
X <- matrix(rnorm(100 * 10), 100, 10)
beta <- c(2, -3, rep(0, 8))
y <- as.vector(X %*% beta + rnorm(100))
norm_result <- normalize(y, X)

Penalized nested Gibbs sampler for Bayesian linear regression

Description

Runs the nested Gibbs sampler for a Gaussian linear model y = X\beta + \epsilon with either a lasso or horseshoe penalty (shrinkage prior) on \beta. The algorithm supports both n \ge p and p > n regimes.

Usage

penalized_nested_Gibbs(
  vy,
  mX,
  penalty_type = c("lasso", "horseshoe"),
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  va_init = NULL,
  verbose = max(1L, floor(as.integer(nsamples)/10)),
  lower = 1e-12,
  upper = 5000,
  s_beta = 1L,
  s_siglam = 1L
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n \times p.

penalty_type

Character string: either "lasso" or "horseshoe".

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Integer number of outer MCMC iterations.

lambda_init

Initial value for \lambda.

va_init

Optional initial values for the local shrinkage parameters (length p). If NULL, a vector of ones is used.

verbose

Print progress every verbose iterations (0 = silent).

lower, upper

Bounds for the slice sampler used for \lambda^2.

s_beta

Integer: number of inner updates of \beta per outer iteration.

s_siglam

Integer: number of inner updates of (\sigma^2,\lambda^2) per \beta update.

Value

A list with components:

mBeta: Matrix of sampled \beta draws (rows correspond to stored draws).
vsigma2: Vector of sampled \sigma^2 draws.
vlambda2: Vector of sampled \lambda^2 draws.

lm_penalized_nested_gibbs().

Examples

set.seed(1)
n <- 50; p <- 10
X <- matrix(rnorm(n * p), n, p)
y <- rnorm(n)

out <- penalized_nested_Gibbs(
  vy = y, mX = X,
  penalty_type = "lasso",
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1,
  va_init = NULL,
  verbose = 0,
  lower = 1e-12,
  upper = 5000,
  s_beta = 1,
  s_siglam = 1
)
str(out)

Penalized PCG sampler: beta block, lambda2 collapsed over sigma2

Description

Partially-collapsed Gibbs (PCG) sampler variant that updates \beta in a dedicated block and samples \lambda^2 using a collapsed step over \sigma^2.

Usage

penalized_pcg_beta_sigma2(
  vy,
  mX,
  penalty_type = c("lasso", "horseshoe"),
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  verbose = max(1L, floor(as.integer(nsamples)/10))
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

penalty_type

Character string: "lasso" or "horseshoe".

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Number of MCMC iterations.

lambda_init

Initial value for \lambda.

sigma2_init

Initial value for \sigma^2.

verbose

Print progress every verbose iterations (0 = silent).

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 40; p <- 6
X <- matrix(rnorm(n * p), n, p)
beta <- c(1.2, 2, -1, 0.5, 0.75, 2.5)
y <- X %*% beta + rnorm(n)

out <- penalized_pcg_beta_sigma2(
  vy = y, mX = X, penalty_type = "horseshoe",
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  verbose = 0
)

summary(out$mBeta)

Penalized PCG sampler: lambda2 collapsed over sigma2

Description

Partially-collapsed Gibbs (PCG) sampler for a Gaussian linear model with a shrinkage prior/penalty on regression coefficients. This variant samples \lambda^2 using a collapsed step over \sigma^2 (see implementation).

Usage

penalized_pcg_lambda2_sigma2(
  vy,
  mX,
  penalty_type = c("lasso", "horseshoe"),
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  verbose = max(1L, floor(as.integer(nsamples)/10))
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

penalty_type

Character string: "lasso" or "horseshoe".

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Number of MCMC iterations.

lambda_init

Initial value for \lambda.

sigma2_init

Initial value for \sigma^2.

verbose

Print progress every verbose iterations (0 = silent).

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 40; p <- 6
X <- matrix(rnorm(n * p), n, p)
beta <- c(1.2, 2, -1, 0.5, 0.75, 2.5)
y <- X %*% beta + rnorm(n)
out <- penalized_pcg_lambda2_sigma2(
  vy = y, mX = X, penalty_type = "lasso",
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200, lambda_init = 1, sigma2_init = 1,
  verbose = 0
)
str(out)

Penalized PCG sampler: sigma2 collapsed over beta

Description

Partially-collapsed Gibbs (PCG) sampler variant that samples \sigma^2 using a collapsed step over \beta. Requires initial values for local scales va_init.

Usage

penalized_pcg_sigma2_beta(
  vy,
  mX,
  penalty_type = c("lasso", "horseshoe"),
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  va_init = NULL,
  verbose = max(1L, floor(as.integer(nsamples)/10)),
  lower = 1e-12,
  upper = 5000
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

penalty_type

Character string: "lasso" or "horseshoe".

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Number of MCMC iterations.

lambda_init

Initial value for \lambda.

sigma2_init

Initial value for \sigma^2.

va_init

Optional initial local-scale vector (length p). If NULL, defaults to rep(1, ncol(mX)).

verbose

Print progress every verbose iterations (0 = silent).

lower, upper

Bounds used by the slice sampler.

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 40; p <- 6
X <- matrix(rnorm(n * p), n, p)
beta <- c(1.2, 2, -1, 0.5, 0.75, 2.5)
y <- X %*% beta + rnorm(n)

out <- penalized_pcg_sigma2_beta(
  vy = y, mX = X, penalty_type = "horseshoe",
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  va_init = rep(1, p),
  verbose = 0
)

summary(out$vsigma2)

Penalized PCG sampler: sigma2 collapsed over lambda2

Description

Partially-collapsed Gibbs (PCG) sampler variant that samples \sigma^2 using a collapsed step over \lambda^2 (see implementation). Requires initial values for local scales va_init; if omitted, it is set to a vector of ones.

Usage

penalized_pcg_sigma2_lambda2(
  vy,
  mX,
  penalty_type = c("lasso", "horseshoe"),
  a,
  b,
  u,
  v,
  nsamples,
  lambda_init = 1,
  sigma2_init = 1,
  va_init = NULL,
  verbose = max(1L, floor(as.integer(nsamples)/10)),
  lower = 1e-12,
  upper = 5000
)

Arguments

vy

Numeric response vector of length n.

mX

Numeric design matrix of dimension n x p.

penalty_type

Character string: "lasso" or "horseshoe".

a, b

Hyperparameters for the inverse-gamma prior on \sigma^2.

u, v

Hyperparameters for the prior on \lambda^2.

nsamples

Number of MCMC iterations.

lambda_init

Initial value for \lambda.

sigma2_init

Initial value for \sigma^2.

va_init

Optional initial local-scale vector (length p). If NULL, defaults to rep(1, ncol(mX)).

verbose

Print progress every verbose iterations (0 = silent).

lower, upper

Bounds used by the slice sampler.

Value

A list with components:

mBeta: Matrix of beta draws (nsamples x p).
vsigma2: Vector of sigma^2 draws (length nsamples).
vlambda2: Vector of lambda^2 draws (length nsamples).

Examples

set.seed(1)
n <- 40; p <- 6
X <- matrix(rnorm(n * p), n, p)
beta <- c(1.2, 2, -1, 0.5, 0.75, 2.5)
y <- X %*% beta + rnorm(n)

out <- penalized_pcg_sigma2_lambda2(
  vy = y, mX = X, penalty_type = "lasso",
  a = 1, b = 1, u = 1, v = 1,
  nsamples = 200,
  lambda_init = 1, sigma2_init = 1,
  va_init = rep(1, p),
  verbose = 0
)

str(out)

Fast High-Dimensional Gibbs Samplers for Bayesian Lasso Regression

Description

Author(s)

See Also

Bayesian lasso Gibbs sampler: 2-block (beta–lambda2) variant

Description

Usage

Arguments

Value

Examples

Bayesian lasso Gibbs sampler: 2-block (beta–sigma2) variant

Description

Usage

Arguments

Value

Examples

Bayesian lasso PCG sampler: lambda2 collapsed over local scales

Description

Usage

Arguments

Value

Examples

Bayesian lasso PCG sampler: sigma2 collapsed over local scales

Description

Usage

Arguments

Value

Examples

Normalize Response and Covariates

Description

Usage

Arguments

Value

Examples

Penalized nested Gibbs sampler for Bayesian linear regression

Description

Usage

Arguments

Value

Examples

Penalized PCG sampler: beta block, lambda2 collapsed over sigma2

Description

Usage

Arguments

Value

Examples

Penalized PCG sampler: lambda2 collapsed over sigma2

Description

Usage

Arguments

Value

Examples

Penalized PCG sampler: sigma2 collapsed over beta

Description

Usage

Arguments

Value

Examples

Penalized PCG sampler: sigma2 collapsed over lambda2

Description

Usage

Arguments

Value

Examples