In statwangz/SuSiE: A simple new approach to variable selection in regression, with application to genetic fine-mapping

This vignette demonstrates the performances of regularized regression methods in the toy example mentioned in Section 2, with a little modification. Here I use two famous regularized regression methods, lasso and EN.

# install.packages(c("glmnet", "ggplot2"))
lapply(c("glmnet", "ggplot2"), require, character.only = TRUE)

Generate data

set.seed(2021)
n <- 300
p <- 100
beta <- rep(0, p)
beta[c(1, 4)] <- 1
X <- matrix(rnorm(n*p), nrow = n, ncol = p)
X[ , 2] <- X[ , 1] + rnorm(n, 0, 0.1)
X[ , 3] <- X[ , 4] + rnorm(n, 0, 0.1)
y <- X %*% beta + rnorm(n)
cor(X[ , 1], X[ , 2])
cor(X[ , 3], X[ , 4])

Normalize

X <- scale(X)
y <- scale(y)

Lasso

fit_lasso <- glmnet(X, y, family = "gaussian", alpha = 1, nlambda = 20)
plot(fit_lasso, xvar = "lambda", label = TRUE)

EN

fit_en <- glmnet(X, y, family = "gaussian", alpha = 0.5, nlambda = 20)
plot(fit_en, xvar= "lambda", label = TRUE)

statwangz/SuSiE documentation built on Nov. 22, 2022, 5:21 p.m.