library(dplyr)

data(pedcan_expression)

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

gene <- sample(colnames(df_expr)[-c(1:2)], size = 1)
x <- df_expr[[gene]]
g <- df_expr$disease
kt <- kruskal.test(x = x, g = g)

test_that("step_select_kruskal works", {

  my_cutoff <- 0.05
  my_correction <- "BH"

  rec <-
    recipe(disease ~ ., data = df_expr) %>%
    step_select_kruskal(all_numeric_predictors(), outcome = "disease",
                        correction = my_correction, cutoff = my_cutoff)

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

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

  expect_equal(nrow(wt_tidy), ncol(df_expr) - 2)
  expect_equal(wt_tidy[wt_tidy$terms == gene, ]$pv, kt$p.value)
  expect_true(all(wt_tidy[wt_tidy$kept, ]$qv <= my_cutoff))
  expect_true(all(wt_tidy[!wt_tidy$kept, ]$qv > my_cutoff))
  expect_equal(p.adjust(wt_tidy$pv, method = my_correction), wt_tidy$qv)


  baked <- bake(prepped, new_data = NULL)

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

})

test_that("step_select_kruskal throw errors", {

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

  expect_error(rec %>%
                 step_select_kruskal(all_numeric_predictors(),
                                      outcome = "event",
                                      correction = c("BH", "fdr"),
                                      cutoff = 0.05) %>%
                 prep(), "`correction` must be a length-one vector.")

  expect_error(rec %>%
                 step_select_kruskal(all_numeric_predictors(),
                                      outcome = "event",
                                      correction = "FDR",
                                      cutoff = 0.05) %>%
                 prep(), '`correction` must be one of "holm", "hochberg", ')

})