test_that("stan execution", {
  skip_if_not_installed("rstanarm")
  skip_on_cran()

  ctrl <- control_parsnip(verbosity = 1, catch = FALSE)
  stan_spec <- poisson_reg() |> set_engine("stan", refresh = 0)

  expect_no_error(
    fit(stan_spec, count ~ ., data = seniors, control = ctrl)
  )
  expect_no_error(
    res <- fit_xy(
      stan_spec,
      x = seniors[, 1:3],
      y = seniors$count,
      control = ctrl
    )
  )

  expect_false(has_multi_predict(res))
  expect_identical(multi_predict_args(res), NA_character_)

  expect_snapshot(error = TRUE, {
    fit(stan_spec, Species ~ term, data = seniors, control = ctrl)
  })
})

test_that("stan prediction", {
  skip_if_not_installed("rstanarm")
  skip_on_cran()

  quiet_ctrl <- control_parsnip(verbosity = 0, catch = TRUE)

  stan_pred <- c(
    6.291754969016,
    6.60564806379581,
    5.64226626722825,
    5.95615936200806,
    4.509666320371
  )

  res_xy <- fit_xy(
    poisson_reg() |> set_engine("stan", seed = 10, chains = 1, refresh = 0),
    x = seniors[, 1:3],
    y = seniors$count,
    control = quiet_ctrl
  )

  expect_identical(
    predict(res_xy, seniors[1:5, 1:3])$.pred,
    stan_pred,
    tolerance = 0.001
  )

  res_form <- fit(
    poisson_reg() |> set_engine("stan", refresh = 0),
    count ~ .,
    data = seniors,
    control = quiet_ctrl
  )
  expect_identical(
    predict(res_form, seniors[1:5, ])$.pred,
    stan_pred,
    tolerance = 0.001
  )
})

test_that("stan intervals", {
  skip_if_not_installed("rstanarm")
  skip_on_cran()

  quiet_ctrl <- control_parsnip(verbosity = 0, catch = TRUE)

  res_xy <- fit_xy(
    poisson_reg() |>
      set_engine("stan", seed = 1333, chains = 10, iter = 1000, refresh = 0),
    x = seniors[, 1:3],
    y = seniors$count,
    control = quiet_ctrl
  )

  confidence_parsnip <-
    predict(res_xy, new_data = seniors[1:5, ], type = "conf_int", level = 0.93)

  prediction_parsnip <-
    predict(res_xy, new_data = seniors[1:5, ], type = "pred_int", level = 0.93)

  ci_lower <- c(
    503.82551692366,
    698.305441192173,
    259.975461883233,
    359.06627830285,
    80.6032220648686
  )
  ci_upper <- c(
    576.810144573584,
    783.482928914321,
    304.974935105559,
    415.426474145917,
    101.122581030063
  )

  pi_lower <- c(483, 676, 244, 340, 71)
  pi_upper <- c(595, 808, 320, 433, 112)

  expect_identical(
    confidence_parsnip$.pred_lower,
    ci_lower,
    tolerance = 1e-2,
    ignore_attr = TRUE
  )
  expect_identical(
    confidence_parsnip$.pred_upper,
    ci_upper,
    tolerance = 1e-2,
    ignore_attr = TRUE
  )

  expect_identical(
    prediction_parsnip$.pred_lower,
    pi_lower,
    tolerance = 1e-2,
    ignore_attr = TRUE
  )
  expect_identical(
    prediction_parsnip$.pred_upper,
    pi_upper,
    tolerance = 1e-2,
    ignore_attr = TRUE
  )
})

test_that("stan intervals with standard error", {
  skip_if_not_installed("rstanarm")
  skip_on_cran()

  quiet_ctrl <- control_parsnip(verbosity = 0, catch = TRUE)

  res_xy <- fit_xy(
    poisson_reg() |>
      set_engine("stan", seed = 1333, chains = 1, iter = 1000, refresh = 0),
    x = seniors[, 1:3],
    y = seniors$count,
    control = quiet_ctrl
  )

  confidence_parsnip <- predict(
    res_xy,
    new_data = seniors[1:5, ],
    type = "conf_int",
    std_error = TRUE
  )
  expect_identical(
    names(confidence_parsnip),
    c(".pred_lower", ".pred_upper", ".std_error")
  )
  expect_identical(nrow(confidence_parsnip), 5L)

  prediction_parsnip <- predict(
    res_xy,
    new_data = seniors[1:5, ],
    type = "pred_int",
    std_error = TRUE
  )
  expect_identical(
    names(prediction_parsnip),
    c(".pred_lower", ".pred_upper", ".std_error")
  )
  expect_identical(nrow(prediction_parsnip), 5L)
})