In jaredhuling/ordinis: Fits penalized regression models with coordinate descent

Introduction to `ordinis`

Fitting penalized regression models

library(ordinis)


set.seed(1)
n <- 100
p <- 1000
m <- 10
b <- matrix(c(runif(m, min = -1), rep(0, p - m)))
x <- matrix(rnorm(n * p, sd = 3), n, p)
y <- drop(x %*% b) + rnorm(n)

# fitting a linear model with an MCP penalty
mod <- ordinis(x, y, 
               penalty = "mcp",
               gamma = 1.4)  # additional tuning param for MCP

plot(mod)

# now applying positivity constraints to parameters and adding a ridge penalty
mod2 <- ordinis(x, y, 
                penalty = "mcp",
                gamma = 1.4,  # additional tuning param for MCP
                lower.limits = rep(0, p), # force all coefficients to be positive
                penalty.factor = c(0, 0, rep(1, p-2)), # don't penalize first two coefficients
                alpha = 0.5)  # use elastic net with alpha = 0.95

plot(mod2)

# use cross validation to select lambda tuning parameter
cvmod <- cv.ordinis(x, y, penalty = "mcp", gamma = 1.4)

plot(cvmod)

# return coefficients with min cv-MSE
coef <- predict(cvmod, type = "coef", s = "lambda.min")

# use cross validation to select lambda tuning parameter
cvmodl1 <- cv.ordinis(x, y, penalty = "lasso")

# return coefficients with min cv-MSE
coefl1 <- predict(cvmodl1, type = "coef", s = "lambda.min")

plot(cvmodl1)

# number selected by MCP
sum(coef[-1] != 0)

# number selected by lasso
sum(coefl1[-1] != 0)

# MCP
round(coef[2:11], 3)

# truth
round(b[1:10], 3)

# lasso
round(coefl1[2:11], 3)