library(dplyr)

data(pedcan_expression)

df_expr <-
  pedcan_expression %>%
  select(event, disease, starts_with("A"))


test_that("step_select_background works", {

  level <- runif(1, min = 1, max = 6)
  prop <- runif(1)
  n <- sample(1:20, size = 1)

  quants <-
    df_expr %>%
    summarize(across(where(is.numeric),
                     ~ unname(quantile(., probs = 1 - prop, type = 1)))) %>%
    unlist()

  rec <-
    recipe(disease ~ ., data = df_expr) %>%
    step_select_background(all_numeric_predictors(),
                           background_level = level, prop_samples = prop)

  expect_equal(nrow(tidy(rec, 1)), 1)

  prepped <- prep(rec)
  bkg_tidy <- tidy(prepped, 1)

  expect_equal(nrow(bkg_tidy), ncol(df_expr) - 2)
  expect_equal(bkg_tidy[bkg_tidy$kept, "terms", drop = TRUE],
               names(quants[quants >= level]))


  baked <- bake(prepped, new_data = NULL)

  expect_setequal(colnames(baked),
                  c(bkg_tidy[bkg_tidy$kept, ]$terms, "event", "disease"))


  kepts <-
    df_expr %>%
    summarize(across(where(is.numeric),
                     ~ sum(. >= level) >= n)) %>%
    unlist()


  bkg_tidy2 <-
    recipe(disease ~ ., data = df_expr) %>%
    step_select_background(all_numeric_predictors(),
                           background_level = level, n_samples = n) %>%
    prep() %>%
    tidy(1)


  expect_equal(bkg_tidy2[bkg_tidy2$kept, "terms", drop = TRUE],
               names(kepts[kepts]))




})


test_that("step_select_background throw errors", {

  rec <-
    recipe(disease ~ ., data = df_expr)

  expect_error(rec %>%
                 step_select_background(all_numeric_predictors(),
                                        n_samples = 10) %>%
                 prep(),
               "background_level")

  expect_error(rec %>%
                 step_select_background(all_numeric_predictors(),
                                        background_level = 2) %>%
                 prep(),
               "n_samples.*prop_samples")

  expect_error(rec %>%
                 step_select_background(all_numeric_predictors(),
                                        background_level = 2, n_samples = 10,
                                        prop_samples = 0.10) %>%
                 prep(),
               "are mutually exclusive")

  expect_error(rec %>%
                 step_select_background(all_predictors(),
                                        background_level = 2, n_samples = 10,
                                        prop_samples = 0.10) %>%
                 prep(),
               "numeric")

})