tests/testthat/test-doJk.R

In sgvignali/SDMtune: Species Distribution Model Selection

skip_on_cran()

v <- c("bio1", "bio12")
t <- SDMtune:::t
t@data <- t@data[, v]
folds <- randomFolds(t, k = 2, only_presence = TRUE)
m <- train("Maxnet", data = t, fc = "lq")
m_cv <- train("Maxnet", data = t, fc = "lq", folds = folds)
files <- list.files(path = file.path(system.file(package = "dismo"), "ex"),
                    pattern = "grd", full.names = TRUE)
predictors <- terra::rast(files)

test_that("The function returns the correct output", {
  # Metric AUC
  # with_only and return models
  jk <- doJk(m,
             metric = "auc",
             variables = v,
             test = t,
             with_only = TRUE,
             return_models = TRUE,
             progress = FALSE)

  expect_type(jk, "list")
  expect_length(jk, 3)
  expect_named(jk, c("results", "models_without", "models_withonly"))
  expect_named(jk$results,
               c("Variable", "Train_AUC_without", "Train_AUC_withonly",
                 "Test_AUC_without", "Test_AUC_withonly"))
  expect_length(jk$results[, 1], 2)
  expect_equal(class(jk$results), "data.frame")
  expect_type(jk$results$Variable, "character")
  expect_length(jk$models_without, 2)
  expect_length(jk$models_withonly, 2)
  expect_named(jk$models_without[[1]]@data@data, "bio12")
  expect_named(jk$models_without[[2]]@data@data, "bio1")
  expect_named(jk$models_withonly[[1]]@data@data, "bio1")
  expect_named(jk$models_withonly[[2]]@data@data, "bio12")

  # with_only = FALSE and return models
  jk <- doJk(m,
             metric = "auc",
             variables = v,
             test = t,
             with_only = FALSE,
             return_models = TRUE,
             progress = FALSE)

  expect_type(jk, "list")
  expect_length(jk, 2)
  expect_named(jk, c("results", "models_without"))

  # with_only and return models = FALSE
  jk <- doJk(m, metric = "auc",
             variables = v,
             test = t,
             with_only = TRUE,
             return_models = FALSE,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "Train_AUC_without", "Train_AUC_withonly",
                     "Test_AUC_without", "Test_AUC_withonly"))
  expect_length(jk[, 1], 2)

  # with_only = FALSE and return models = FALSE
  jk <- doJk(m,
             metric = "auc",
             variables = v,
             test = t,
             with_only = FALSE,
             return_models = FALSE,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "Train_AUC_without", "Test_AUC_without"))
  expect_length(jk[, 1], 2)

  # cross validation model, with_only = FALSE and return models = FALSE
  jk <- doJk(m_cv,
             metric = "auc",
             variables = v,
             test = t,
             with_only = FALSE,
             return_models = FALSE,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "Train_AUC_without", "Test_AUC_without"))
  expect_length(jk[, 1], 2)
  expect_type(jk$Variable, "character")

  # all variables, with_only = FALSE and return models = FALSE
  jk <- doJk(m,
             metric = "auc",
             test = t,
             with_only = FALSE,
             return_models = FALSE,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "Train_AUC_without", "Test_AUC_without"))
  expect_length(jk[, 1], 2)
  expect_type(jk$Variable, "character")

  # Metric TSS
  # with_only and return models = FALSE
  jk <- doJk(m,
             metric = "tss",
             variables = v,
             test = t,
             with_only = TRUE,
             return_models = FALSE,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "Train_TSS_without", "Train_TSS_withonly",
                     "Test_TSS_without", "Test_TSS_withonly"))
  expect_length(jk[, 1], 2)

  # test, with_only = FALSE and return models = FALSE
  jk <- doJk(m,
             metric = "tss",
             variables = v,
             test = t,
             with_only = FALSE,
             return_models = FALSE,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "Train_TSS_without", "Test_TSS_without"))
  expect_length(jk[, 1], 2)

  # Metric AICc
  # with_only and return models = FALSE
  jk <- doJk(m,
             metric = "aicc",
             variables = v,
             with_only = TRUE,
             return_models = FALSE,
             env = predictors,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "AICc_without", "AICc_withonly"))
  expect_length(jk[, 1], 2)

  # test, with_only = FALSE and return models = FALSE
  jk <- doJk(m,
             metric = "aicc",
             variables = v,
             with_only = FALSE,
             return_models = FALSE,
             env = predictors,
             progress = FALSE)

  expect_equal(class(jk), "data.frame")
  expect_named(jk, c("Variable", "AICc_without"))
  expect_length(jk[, 1], 2)
})

test_that("The function runs without test dataset", {
  expect_error(doJk(m,
                    metric = "auc",
                    variables = "bio1",
                    with_only = FALSE,
                    progress = FALSE),
               NA)
})

test_that("The function raises errors", {
  expect_snapshot_error(doJk(m,
                             metric = "aicc",
                             variables = v,
                             env = "spam",
                             progress = FALSE))
})

# TODO: Remove with version 2.0.0
test_that("The function raises an error if a raster object is used", {
  class(predictors) <- "Raster"
  expect_snapshot_error(doJk(m,
                             metric = "aicc",
                             variables = v,
                             env = predictors,
                             test = NULL,
                             with_only = FALSE,
                             return_models = FALSE,
                             progress = FALSE))
})