tests/testthat/test-acquisition.R

In tune: Tidy Tuning Tools

test_that("conf_bound interface", {
  expect_snapshot(error = TRUE, conf_bound("a"))
  expect_snapshot(error = TRUE, conf_bound(function() 1))

  test_res <- tibble::tibble(.mean = 1:10, .sd = c((1:9) / 10, NA_real_))
  expect_snapshot(
    error = TRUE,
    predict(conf_bound(), test_res, maximize = 2, iter = 1)
  )
})

test_that("conf_bound calculations", {
  test_res <- tibble::tibble(.mean = 1:10, .sd = c((1:9) / 10, NA_real_))
  dbled <- function(x) x^2

  expect_equal(
    predict(conf_bound(kappa = 1), test_res, maximize = TRUE, iter = 1),
    test_res %>% mutate(objective = .mean + 1 * .sd) %>% select(objective)
  )
  expect_equal(
    predict(conf_bound(2), test_res, maximize = TRUE, iter = 1),
    test_res %>% mutate(objective = .mean + 2 * .sd) %>% select(objective)
  )
  expect_equal(
    predict(conf_bound(dbled), test_res, maximize = TRUE, iter = 2),
    test_res %>% mutate(objective = .mean + 4 * .sd) %>% select(objective)
  )
  expect_equal(
    predict(conf_bound(kappa = 1), test_res, maximize = FALSE, iter = 1),
    test_res %>% mutate(objective = -(.mean + 1 * .sd)) %>% select(objective)
  )
})


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

test_that("prob_improve interface", {
  test_res <- tibble::tibble(.mean = 1:10, .sd = c((1:9) / 10, NA_real_))

  expect_snapshot(error = TRUE, prob_improve("a"))
  expect_snapshot(error = TRUE, prob_improve(function() 1))
  expect_snapshot(error = TRUE, predict(prob_improve(), test_res, maximize = 2, iter = 1))
  expect_snapshot(error = TRUE, predict(prob_improve(), test_res, maximize = TRUE, iter = 1, best = NA))
  expect_snapshot(error = TRUE, predict(prob_improve(), test_res, maximize = TRUE, iter = 1, best = "WAT"))
})

test_that("prob_improve calculations", {
  test_res <- tibble::tibble(.mean = 1:10, .sd = c((1:9) / 10, NA_real_))
  dbled <- function(x) x^2

  expect_equal(
    predict(prob_improve(), test_res, maximize = TRUE, iter = 1, best = 15),
    test_res %>%
      mutate(objective = pnorm((.mean - 15) / .sd)) %>%
      select(objective)
  )
  expect_equal(
    predict(prob_improve(), test_res, maximize = FALSE, iter = 1, best = -2),
    test_res %>%
      mutate(objective = pnorm((-2 - .mean) / .sd)) %>%
      select(objective)
  )

  expect_equal(
    predict(prob_improve(.1), test_res, maximize = TRUE, iter = 1, best = 15),
    test_res %>%
      mutate(objective = pnorm((.mean - 15 - .1) / .sd)) %>%
      select(objective)
  )
  expect_equal(
    predict(prob_improve(.1), test_res, maximize = FALSE, iter = 1, best = -2),
    test_res %>%
      mutate(objective = pnorm((-2 + .1 - .mean) / .sd)) %>%
      select(objective)
  )

  expect_equal(
    predict(prob_improve(dbled), test_res, maximize = TRUE, iter = 2, best = 15),
    test_res %>%
      mutate(objective = pnorm((.mean - 15 - 4) / .sd)) %>%
      select(objective)
  )
  expect_equal(
    predict(prob_improve(dbled), test_res, maximize = FALSE, iter = 4, best = -2),
    test_res %>%
      mutate(objective = pnorm((-2 + 16 - .mean) / .sd)) %>%
      select(objective)
  )
})


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

test_that("exp_improve interface", {
  test_res <- tibble::tibble(.mean = 1:10, .sd = c((1:9) / 10, NA_real_))

  expect_snapshot(error = TRUE, exp_improve("a"))
  expect_snapshot(error = TRUE, exp_improve(function() 2))
  expect_snapshot(error = TRUE, predict(exp_improve(), test_res, maximize = 2, iter = 1))
  expect_snapshot(error = TRUE, predict(exp_improve(), test_res, maximize = TRUE, iter = 1, best = NA))
  expect_snapshot(error = TRUE, predict(exp_improve(), test_res, maximize = TRUE, iter = 1, best = "WAT"))
})

test_that("exp_improve calculations", {
  test_res <- tibble::tibble(.mean = 1:10, .sd = c((1:9) / 10, NA_real_))
  dbled <- function(x) x^2

  expect_equal(
    predict(exp_improve(), test_res, maximize = TRUE, iter = 1, best = 15),
    test_res %>%
      mutate(
        diff = .mean - 15,
        objective = (diff * pnorm(diff / .sd)) + (.sd * dnorm(diff / .sd))
      ) %>%
      select(objective)
  )
  expect_equal(
    predict(exp_improve(), test_res, maximize = FALSE, iter = 1, best = 15),
    test_res %>%
      mutate(
        diff = 15 - .mean,
        objective = (diff * pnorm(diff / .sd)) + (.sd * dnorm(diff / .sd))
      ) %>%
      select(objective)
  )

  expect_equal(
    predict(exp_improve(1), test_res, maximize = TRUE, iter = 1, best = 15),
    test_res %>%
      mutate(
        diff = .mean - 16,
        objective = (diff * pnorm(diff / .sd)) + (.sd * dnorm(diff / .sd))
      ) %>%
      select(objective)
  )
  expect_equal(
    predict(exp_improve(1), test_res, maximize = FALSE, iter = 1, best = 15),
    test_res %>%
      mutate(
        diff = 16 - .mean,
        objective = (diff * pnorm(diff / .sd)) + (.sd * dnorm(diff / .sd))
      ) %>%
      select(objective)
  )

  expect_equal(
    predict(exp_improve(dbled), test_res, maximize = TRUE, iter = 2, best = 15),
    test_res %>%
      mutate(
        diff = .mean - 19,
        objective = (diff * pnorm(diff / .sd)) + (.sd * dnorm(diff / .sd))
      ) %>%
      select(objective)
  )
  expect_equal(
    predict(exp_improve(dbled), test_res, maximize = FALSE, iter = 2, best = 15),
    test_res %>%
      mutate(
        diff = 19 - .mean,
        objective = (diff * pnorm(diff / .sd)) + (.sd * dnorm(diff / .sd))
      ) %>%
      select(objective)
  )
})