In stephenslab/susieR: Sum of Single Effects Linear Regression

knitr::opts_chunk$set(echo = TRUE)

The goal here is to test the function susie_auto which tries to make susie run well even in some tricky situations (eg where $L$ is big, which susie is not well suited to).

This is an example Lei Sun showed us from the paper demonstrating that false positives occur early on the Lasso path. Here I change L to 20 instead of 200 to make it run fast to begin with.

set.seed(777)
library(susieR)
L <- 20
X <- matrix(rnorm(1010 * 1000), 1010, 1000)
beta <- rep(0, 1000)
beta[1 : L] <- 100
y <- X %*% beta + rnorm(1010)

s <- susie_auto(X,y,verbose=TRUE)
s$sa2

Now try L=200

set.seed(777)
L <- 200
X <- matrix(rnorm(1010 * 1000), 1010, 1000)
beta <- rep(0, 1000)
beta[1 : L] <- 100
y <- X %*% beta + rnorm(1010)
s2 <- susie_auto(X,y,verbose=TRUE)
s2$sa2