tests/testthat/test-alpha-scaling.R

In SLOPE: Sorted L1 Penalized Estimation

test_that("penalty strength is invariant to number of observations", {
  x <- abalone$x
  y <- abalone$y

  x_small <- x
  y_small <- y
  x_large <- do.call(rbind, rep(list(x), 2))
  y_large <- rep(y, 2)

  for (family in c("gaussian", "poisson")) {
    for (scale in c("none", "sd", "l1", "l2")) {
      alpha <- switch(scale,
        none = 0.1,
        sd = 0.3,
        l1 = 0.5,
        l2 = 0.3
      )

      for (center in c(TRUE, FALSE)) {
        f0 <- SLOPE(
          x_small,
          y_small,
          family = family,
          alpha = alpha,
          scale = scale,
          center = center,
          tol_rel_gap = 1e-6,
          tol_infeas = 1e-6
        )
        f1 <- SLOPE(
          x_large,
          y_large,
          family = family,
          alpha = alpha,
          scale = scale,
          center = center,
          tol_rel_gap = 1e-6,
          tol_infeas = 1e-6
        )

        expect_equivalent(coef(f0), coef(f1), tol = 1e-3)
      }
    }
  }
})