README.md

In antshi/constrLasso: Contrained lasso regression and solution path in R as in [Gaines et al. (2018)]

constrlasso

The package constrlasso includes functions for a constrained lasso regression and a solution path algorithm as in Gaines et al. (2018).

Installation

You can install this version of the package from github with:

install.packages("devtools")
library(devtools)
install_github("antshi/constrlasso")
library(constrlasso)

Constrained Lasso Regression

These are basic examples which show you how to use the function constrlasso. First, let's prepare with

# include the package
library(constrlasso)

# generate some data
set.seed(1234)
n <- 200 # number of observations
p <- 150 # number of regressors
real_p <- 50 # number of true predictors
Xmat <- matrix(rnorm(n * p), nrow = n, ncol = p)
yvec <- apply(Xmat[, 1:real_p], 1, sum) + rnorm(n)

Example 1

No constraints and no penalty.

results_reg_no_constr <- constrlasso(Xmat, yvec, lambda = 0)

Example 2

Included constraints and penalty.

# build constraints

# equality constraints (sum constraint)
Aeq <- matrix(1, 1, p)
beq <- matrix(1, 1, 1)

# inequality constraints (upper and lower bounds set to 0.5 and -0.5, respectively)
A1 <- matrix(diag(1, p), p, p)
b1 <- matrix(0.5, p, 1)
A2 <- matrix(-diag(1, p), p, p)
b2 <- matrix(0.5, p, 1)
A <- rbind(A1, A2)
b <- rbind(b1, b2)

results_reg_constr <- constrlasso(Xmat, yvec, Aeq = Aeq, beq = beq, lambda = 3)
sum(results_reg_constr[[1]]) # should be equal to 1

results_reg_constr2 <- constrlasso(Xmat, yvec, Aeq = Aeq, beq = beq, A = A, b = b, lambda = 2)
sum(results_reg_constr2[[1]]) # should be equal to 1
which(abs(results_reg_constr2[[1]]) > 0.5) # should not contain entries

Constrained Lasso Solution Path

These are basic examples which show you how to use the function constrlasso_path. First, let's prepare with

# include the package
library(constrlasso)

# generate some data
set.seed(1234)
n <- 200 # number of observations
p <- 150 # number of regressors
real_p <- 50 # number of true predictors
Xmat <- matrix(rnorm(n * p), nrow = n, ncol = p)
yvec <- apply(Xmat[, 1:real_p], 1, sum) + rnorm(n)

Example 3

results_path <- constrlasso_path(Xmat, yvec)

Example 4

# build constraints

# equality constraints (sum constraint)
Aeq <- matrix(1, 1, p)
beq <- matrix(1, 1, 1)

# inequality constraints (upper and lower bounds set to 0.5 and -0.5, respectively)
A1 <- matrix(diag(1, p), p, p)
b1 <- matrix(0.5, p, 1)
A2 <- matrix(-diag(1, p), p, p)
b2 <- matrix(0.5, p, 1)
A <- rbind(A1, A2)
b <- rbind(b1, b2)

results_path_constr <- constrlasso_path(Xmat, yvec, Aeq = Aeq, beq = beq)
apply(results_path_constr[[1]], 2, sum) # should be equal to 1 along the path


results_path_constr2 <- constrlasso_path(Xmat, yvec, Aeq = Aeq, beq = beq, A = A, b = b)
apply(results_path_constr2[[1]], 2, sum) # should be equal to 1 along the path
apply(results_path_constr2[[1]], 2, function(x) which(abs(round(x, 10)) > 0.5)) # should not contain entries

antshi/constrLasso documentation built on April 12, 2025, 9:39 a.m.