tests/testthat/test-dCVnet_coef.R

In AndrewLawrence/dCVnet: doubly cross-validated elastic-net regularised generalised linear models

# see make_testdata.R for data/model.RData

load(file = "data/model.RData")
# m

# there are 30 predictors plus an intercept
#   evaluated for 2x10-fold repeated CV
# Thus 31 x 20 =

# check dims:
test_that("Coef produces the expected dimensions", {
  types_of_coefs <- c("production",
                      "all",
                      "mean",
                      "median",
                      "byrep",
                      "byrep_mean",
                      "byrep_median")
  names(types_of_coefs) <- types_of_coefs

  coef_list <- lapply(types_of_coefs, function(x) coef(m, type = x))

  coef_nrow <- vapply(coef_list, nrow, FUN.VALUE = 1L)
  coef_ncol <- vapply(coef_list, ncol, FUN.VALUE = 1L)

  expected_nrow <- 31L * c(1L, 20L, 1L, 1L, 2L, 1L, 1L)
  expected_ncol <- c(2L, 4L, 2L, 2L, 3L, 2L, 2L)
  names(expected_ncol) <- names(expected_nrow) <- types_of_coefs

  # Check coefficients are numeric:
  coef_classes <-
    vapply(coef_list, function(x) class(x[["Coef"]]), FUN.VALUE = "chr")

  # check rows:
  expect_equal(coef_nrow, expected_nrow)
  # check cols:
  expect_equal(coef_ncol, expected_ncol)

  # "nonsensical coef types produce errors",
  expect_error(coef(m, type = "hafdsa"))

  # "coefficients are numeric", {
  expect_equal(coef_classes[[1]], "numeric")
  expect_equal(length(unique(coef_classes)), 1L)
})