tests/testthat/test-impute_linear.R

In recipes: Preprocessing and Feature Engineering Steps for Modeling

library(testthat)
skip_if_not_installed("modeldata")

data(ames, package = "modeldata")

ames_dat <- ames %>%
  select(Lot_Frontage, Lot_Area, Total_Bsmt_SF) %>%
  mutate(Lot_Frontage = na_if(Lot_Frontage, 0)) %>%
  mutate(Total_Bsmt_SF = na_if(Total_Bsmt_SF, 0))

tg_dat <- ToothGrowth %>%
  mutate(dose = na_if(dose, 0.5))


test_that("Does the imputation (no NAs), and does it correctly.", {
  missing_ind <- which(is.na(ames_dat$Lot_Frontage), arr.ind = TRUE)

  imputed <- recipe(head(ames_dat)) %>%
    step_impute_linear(Lot_Frontage, impute_with = c("Lot_Area")) %>%
    prep(ames_dat) %>%
    bake(new_data = NULL) %>%
    pull(Lot_Frontage) %>%
    .[missing_ind]

  lm_predicted <- lm(Lot_Frontage ~ Lot_Area, data = ames_dat) %>%
    predict(newdata = ames_dat[missing_ind, ]) %>%
    unname()

  expect_equal(imputed, lm_predicted)
  expect_equal(sum(is.na(imputed)), 0)
})

test_that("All NA values", {
  imputed <- recipe(head(ames_dat)) %>%
    step_impute_linear(Lot_Frontage, impute_with = c("Lot_Area")) %>%
    prep(ames_dat)

  imputed_te <- bake(imputed, ames_dat %>% mutate(Lot_Frontage = NA))

  lm_predicted <- lm(Lot_Frontage ~ Lot_Area, data = ames_dat) %>%
    predict(newdata = ames_dat) %>%
    unname()

  expect_equal(unname(imputed_te$Lot_Frontage), lm_predicted)
  expect_equal(sum(is.na(imputed_te$Lot_Frontage)), 0)
})


test_that("Returns correct models.", {
  imputed <- recipe(head(ames_dat)) %>%
    step_impute_linear(Lot_Frontage, Total_Bsmt_SF, impute_with = c("Lot_Area")) %>%
    prep(ames_dat)

  expect_equal(length(imputed$steps[[1]]$models), 2)
  expect_equal(
    sort(names(imputed$steps[[1]]$models)),
    c("Lot_Frontage", "Total_Bsmt_SF")
  )
})

test_that("Fails when one of the variables to impute is non-numeric.", {
  expect_snapshot(error = TRUE,
    recipe(tg_dat) %>%
      step_impute_linear(supp, impute_with = c("len")) %>%
      prep(tg_dat)
  )
  expect_snapshot(error = TRUE,
    recipe(tg_dat) %>%
      step_impute_linear(supp, dose, impute_with = c("len")) %>%
      prep(tg_dat)
  )
})


test_that("Maintain data type", {
  ames_integer <- ames
  ames_integer$TotRms_AbvGrd[1:10] <- NA_integer_
  integer_rec <- recipe(~., data = ames_integer) %>%
    step_impute_linear(TotRms_AbvGrd, impute_with = vars(Bedroom_AbvGr, Gr_Liv_Area)) %>%
    prep()
  expect_true(
    is.integer(bake(integer_rec, ames_integer, TotRms_AbvGrd) %>% pull(TotRms_AbvGrd))
  )
})

test_that("case weights", {
  missing_ind <- which(is.na(ames_dat$Lot_Frontage), arr.ind = TRUE)

  ames_dat_cw <- ames_dat %>%
    mutate(Total_Bsmt_SF = frequency_weights(Total_Bsmt_SF))

  rec_prepped <- recipe(head(ames_dat_cw)) %>%
    step_impute_linear(Lot_Frontage, impute_with = c("Lot_Area")) %>%
    prep(ames_dat_cw)

  imputed <- rec_prepped %>%
    bake(new_data = NULL) %>%
    pull(Lot_Frontage) %>%
    .[missing_ind]

  lm_predicted <- lm(Lot_Frontage ~ Lot_Area, data = ames_dat,
                     weights = Total_Bsmt_SF) %>%
    predict(newdata = ames_dat[missing_ind, ]) %>%
    unname()

  expect_equal(imputed, lm_predicted)
  expect_equal(sum(is.na(imputed)), 0)

  expect_snapshot(rec_prepped)

  # ----------------------------------------------------------------------------

  missing_ind <- which(is.na(ames_dat$Lot_Frontage), arr.ind = TRUE)

  ames_dat_cw <- ames_dat %>%
    mutate(Total_Bsmt_SF = importance_weights(Total_Bsmt_SF))

  rec_prepped <- recipe(head(ames_dat_cw)) %>%
    step_impute_linear(Lot_Frontage, impute_with = c("Lot_Area")) %>%
    prep(ames_dat_cw)

  imputed <- rec_prepped %>%
    bake(new_data = NULL) %>%
    pull(Lot_Frontage) %>%
    .[missing_ind]

  lm_predicted <- lm(Lot_Frontage ~ Lot_Area, data = ames_dat) %>%
    predict(newdata = ames_dat[missing_ind, ]) %>%
    unname()

  expect_equal(imputed, lm_predicted)
  expect_equal(sum(is.na(imputed)), 0)

  expect_snapshot(rec_prepped)
})

# Infrastructure ---------------------------------------------------------------

test_that("bake method errors when needed non-standard role columns are missing", {
  rec <- recipe(head(ames_dat)) %>%
    step_impute_linear(Lot_Frontage, impute_with = c("Lot_Area")) %>%
    update_role(Lot_Frontage, new_role = "potato") %>%
    update_role_requirements(role = "potato", bake = FALSE) %>%
    prep(ames_dat)

  expect_error(bake(rec, new_data = ames_dat[, 2:3]),
               class = "new_data_missing_column")
})

test_that("empty printing", {
  rec <- recipe(mpg ~ ., mtcars)
  rec <- step_impute_linear(rec)

  expect_snapshot(rec)

  rec <- prep(rec, mtcars)

  expect_snapshot(rec)
})

test_that("empty selection prep/bake is a no-op", {
  rec1 <- recipe(mpg ~ ., mtcars)
  rec2 <- step_impute_linear(rec1)

  rec1 <- prep(rec1, mtcars)
  rec2 <- prep(rec2, mtcars)

  baked1 <- bake(rec1, mtcars)
  baked2 <- bake(rec2, mtcars)

  expect_identical(baked1, baked2)
})

test_that("empty selection tidy method works", {
  rec <- recipe(mpg ~ ., mtcars)
  rec <- step_impute_linear(rec)

  expect <- tibble(terms = character(), model = list(), id = character())

  expect_identical(tidy(rec, number = 1), expect)

  rec <- prep(rec, mtcars)

  expect_identical(tidy(rec, number = 1), expect)
})

test_that("printing", {
  rec <- recipe(ames_dat) %>%
    step_impute_linear(Lot_Frontage, impute_with = imp_vars(Lot_Area))

  expect_snapshot(print(rec))
  expect_snapshot(prep(rec))
})