tests/testthat/test-linear_reg-lme4.R

In tidymodels/multilevelmod: Model Wrappers for Multi-Level Models

test_that('lme4 execution', {
  skip_if_not_installed("lme4")
  skip_on_cran()

  lme4_cl <- call2("lmer", .ns = "lme4", f, data = expr(riesby_tr))

  set.seed(2452)
  lme4_mod <- eval_tidy(lme4_cl)

  set.seed(2452)
  expect_error(
    ps_mod <-
      linear_reg() %>%
      set_engine("lmer") %>%
      fit(f, data = riesby_tr),
    regex = NA
  )

  expect_equal(
    coef(ps_mod$fit)$subject,
    coef(lme4_mod)$subject
  )

  lme4_pred <- predict(lme4_mod, riesby_te, allow.new.levels = TRUE)

  ps_pred <- predict(ps_mod, riesby_te)

  expect_equal(unname(lme4_pred), ps_pred$.pred)

})


test_that('mode specific package dependencies', {
  expect_identical(
    get_from_env(paste0("linear_reg", "_pkgs")) %>%
      dplyr::filter(engine == "lmer", mode == "classification") %>%
      dplyr::pull(pkg),
    list()
  )

  expect_identical(
    get_from_env(paste0("linear_reg", "_pkgs")) %>%
      dplyr::filter(engine == "lmer", mode == "regression") %>%
      dplyr::pull(pkg),
    list(c("lme4", "multilevelmod"))
  )
})


test_that('random interactions and/or nesting', {
  skip_if_not_installed("lme4")
  skip_if_not_installed("ggplot2")
  skip_on_cran()

  data(mpg, package = "ggplot2")

  fit <-
    linear_reg() %>%
    set_engine("lmer") %>%
    fit(cty ~ year + (1|manufacturer/model), data = mpg)

  expect_error(predict(fit, new_data = mpg), regexp = NA)
  # Now predict with new levels for the random effect columns, esp
  # when there is nesting/interactions
  expect_error(
    predict(fit, new_data = tibble::tibble(year = 2000, model = "dan", manufacturer = "max")),
    regexp = NA
  )



})