This fork of the {PGEE} R package is modified from the original by adding the option of applying a ridge /
You can install the development version of PGEE from GitHub with:
# install.packages("devtools")
devtools::install_github("ad1729/PGEE")The ridge penalty can be specified by passing the penalty_type = "ridge" argument to the PGEE() call. By default, penalty_type = "SCAD".
library(PGEE)
# library(dplyr)
n <- 1000
set.seed(5)
dat <- dplyr::tibble(
exposure = rbinom(n, size = 1, prob = 0.6),
# adding a small factor to the time runif endpoints here to avoid spline warnings
# regarding predictions at boundary knots
time = round(runif(n, min = -3.05, max = 12.05), 2),
#time = sample(x = -4:13, size = n, replace = TRUE),
logodds = -0.1 + (-0.2 * (time ^ 2)) + (0.7 * exposure),
response = rbinom(n, size = 1, prob = plogis(logodds)),
id = factor(rep(LETTERS[1:10], each = 100))
) %>%
dplyr::mutate(exposure = factor(exposure, levels = c(0, 1))) %>%
dplyr::arrange(id)
# pindex = c(1, 2) indicates that the intercept and exposure terms should not be penalized
mod <- PGEE(
response ~ exposure + splines2::bSpline(time, knots = seq(-2, 10, 1)),
data = dat, family = "binomial", corstr = "exchangeable", id = id,
lambda = 0.001, scale.fix = TRUE, pindex = c(1, 2), silent = FALSE,
penalty_type = "ridge"
)
mod
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent