tests/testthat/test_ris_cv.R

In pense: Penalized Elastic Net S/MM-Estimator of Regression

library(pense)
library(testthat)

test_that("RIS-CV", {
  skip_if_not(nzchar(Sys.getenv('PENSE_TEST_FULL')),
              message = 'Environment variable `PENSE_TEST_FULL` not defined.')

  ncores <- if(isTRUE(pense:::.k_multithreading_support)) {
    2L
  } else {
    1L
  }

  n <- 50L
  p <- 10L
  nlambda <- 25L
  alphas <- c(0.1, 0.8)

  set.seed(123)
  x <- matrix(rcauchy(n * p), ncol = p)
  y <- 2 + rowSums(x[, 1:5]) / 5 + rnorm(n, sd = 4)

  max_solutions <- 10

  expect_no_condition(
    pr <- pense_cv(x, y,
                   cv_k = 3, cv_repl = 2,
                   cv_type = 'ris',
                   alpha = alphas,
                   nlambda = nlambda,
                   nlambda_enpy = 5,
                   max_solutions = max_solutions,
                   ncores = ncores,
                   bdp = 0.25,
                   sparse = FALSE,
                   eps = 1e-8,
                   enpy_opts = enpy_options(retain_max = 5, en_algorithm_opts = en_lars_options()),
                   algorithm_opts = mm_algorithm_options(en_algorithm_opts = en_lars_options()))
  )

  expect_length(pr$estimates, length(alphas) * nlambda)
  expect_contains(colnames(pr$cvres), c('cvavg', 'cvse', 'lambda', 'alpha'))
  expect_length(coef(pr), p + 1L)
})

test_that("RIS-CV (optimal rho)", {
  skip_if_not(nzchar(Sys.getenv('PENSE_TEST_FULL')),
              message = 'Environment variable `PENSE_TEST_FULL` not defined.')

  ncores <- if(isTRUE(pense:::.k_multithreading_support)) {
    2L
  } else {
    1L
  }

  n <- 50L
  p <- 10L
  nlambda <- 25L
  alphas <- c(0.1, 0.8)

  set.seed(123)
  x <- matrix(rcauchy(n * p), ncol = p)
  y <- 2 + rowSums(x[, 1:5]) / 5 + rnorm(n, sd = 4)

  max_solutions <- 10

  expect_no_condition(
    pr <- pense_cv(x, y,
                   cv_k = 3, cv_repl = 2,
                   cv_type = 'ris',
                   alpha = alphas,
                   nlambda = nlambda,
                   nlambda_enpy = 5,
                   max_solutions = max_solutions,
                   ncores = ncores,
                   bdp = 0.25,
                   sparse = FALSE,
                   eps = 1e-8,
                   mscle_opts = mscale_algorithm_options("mopt"),
                   enpy_opts = enpy_options(retain_max = 5, en_algorithm_opts = en_lars_options()),
                   algorithm_opts = mm_algorithm_options(en_algorithm_opts = en_lars_options()))
  )

  expect_length(pr$estimates, length(alphas) * nlambda)
  expect_contains(colnames(pr$cvres), c('cvavg', 'cvse', 'lambda', 'alpha'))
  expect_length(coef(pr), p + 1L)
})