data-raw/sim_performance_np_ratio.R

In jolars/strong.SLOPE.simulations: Data and Source Code for Simulations for the SLOPE Strong Rule Paper

# simulations for testing performance of screening rule as ratio between n and p vary

library(SLOPE)

n <- 1000

out <- data.frame()

n_it <- 10
n_ratio <- 10
min_ratio <- 0.01
max_ratio <- 10
ratio <- signif(exp(seq(log(min_ratio),
                        log(max_ratio),
                        length.out = n_ratio)), 1)
q <- 0.1

screening <- c(TRUE, FALSE)

set.seed(1044)

iter <- 0

for (j in seq_along(ratio)) {
  p <- n*ratio[j]
  k <- floor(p*q)

  for (i in seq_len(n_it)) {
    x <- matrix(rnorm(n*p), n)
    beta <- double(p)
    beta[1:k] <- sample(c(2, -2), k, replace = TRUE)

    y <- x %*% beta + rnorm(n)

    for (k in 1:length(screening)) {
      iter <- iter + 1

      cat("iter:", iter, "/", n_it*length(ratio)*2, "\t",
          "ratio:", ratio[j], "\t",
          "screening:", screening[k], "\n")

      time <- system.time({
        fit <- SLOPE(x,
                     y,
                     lambda = "bh",
                     q = q,
                     intercept = FALSE,
                     screen = screening[k])
      })

      tmp <- data.frame(n = n,
                        p = p,
                        ratio = ratio[j],
                        screening = screening[k],
                        time = time[3])
      out <- rbind(out, tmp)
    }
  }
}

sim_performance_np_ratio <- out
overwrite <- file.exists("data/sim_performance_np_ratio.rda")
usethis::use_data(sim_performance_np_ratio, overwrite = overwrite)