## =============================================================================
## arimasel tests: modernization features (seasonal, xreg, CV, seasonal diagnostics)
## =============================================================================

context("cart_arima seasonal search")

test_that("cart_arima with seasonal argument returns seasonal model strings", {
  res <- cart_arima(.x_seas, p_set = 0:1, d_set = 0L, q_set = 0:1,
                    seasonal = list(P = 0:1, D = 0L, Q = 0:1, period = 12))
  expect_s3_class(res, "cartARIMA")
  expect_true(all(grepl("\\[12\\]$", res$full_table$Model)))
})

test_that("cart_arima seasonal n_total equals full seasonal cartesian product", {
  res <- cart_arima(.x_seas, p_set = 0:1, d_set = 0L, q_set = 0:1,
                    seasonal = list(P = 0:1, D = 0L, Q = 0:1, period = 12))
  expect_equal(res$n_total, 2L * 1L * 2L * 2L * 1L * 2L)
})

test_that("cart_arima errors on invalid seasonal period", {
  expect_error(
    cart_arima(.x_seas, p_set = 0:1, d_set = 0L, q_set = 0:1,
              seasonal = list(period = 1L)),
    "period"
  )
})

test_that("cart_arima without seasonal still works (backward compatible)", {
  res <- cart_arima(.x_stat, p_set = 0:1, d_set = 0L, q_set = 0:1)
  expect_null(res$seasonal)
})

## --------------------------------------------------------------------------- #
context("cart_arima xreg support")

test_that("cart_arima accepts xreg and stores it", {
  res <- cart_arima(.x_xreg_y, p_set = 0:1, d_set = 0L, q_set = 0:1,
                    xreg = .x_xreg_z)
  expect_s3_class(res, "cartARIMA")
  expect_false(is.null(res$xreg))
})

test_that("cart_arima errors when xreg length mismatches x", {
  expect_error(
    cart_arima(.x_xreg_y, p_set = 0:1, d_set = 0L, q_set = 0:1,
              xreg = .x_xreg_z[1:10, , drop = FALSE]),
    "same number of rows"
  )
})

test_that("arima_forecast requires newxreg when model was fit with xreg", {
  res <- cart_arima(.x_xreg_y, p_set = 0:1, d_set = 0L, q_set = 0:1,
                    xreg = .x_xreg_z)
  expect_error(arima_forecast(res, h = 3L, plot = FALSE), "newxreg")
})

test_that("arima_forecast with valid newxreg returns forecasts of length h", {
  res <- cart_arima(.x_xreg_y, p_set = 0:1, d_set = 0L, q_set = 0:1,
                    xreg = .x_xreg_z)
  new_z <- matrix(rnorm(3), ncol = 1)
  fc <- arima_forecast(res, h = 3L, newxreg = new_z, plot = FALSE)
  expect_length(fc$mean, 3L)
})

## --------------------------------------------------------------------------- #
context("arima_forecast prediction interval correctness")

test_that("arima_forecast lower is always <= upper (regression test)", {
  res <- cart_arima(.x_stat, p_set = 0:1, d_set = 0L, q_set = 0:1)
  fc  <- arima_forecast(res, h = 6L, plot = FALSE, level = c(50, 80, 95, 99))
  expect_true(all(fc$lower <= fc$upper))
})

test_that("wider confidence levels produce wider intervals", {
  res <- cart_arima(.x_stat, p_set = 0:1, d_set = 0L, q_set = 0:1)
  fc  <- arima_forecast(res, h = 5L, plot = FALSE, level = c(80, 95))
  width_80 <- fc$upper[, "80%"] - fc$lower[, "80%"]
  width_95 <- fc$upper[, "95%"] - fc$lower[, "95%"]
  expect_true(all(width_95 >= width_80))
})

## --------------------------------------------------------------------------- #
context("fitted.cartARIMA (regression test)")

test_that("fitted() returns a finite numeric vector", {
  res <- cart_arima(.x_stat, p_set = 0:1, d_set = 0L, q_set = 0:1)
  fv  <- fitted(res)
  expect_true(is.numeric(fv))
  expect_true(all(is.finite(fv)))
  expect_equal(length(fv), res$n_obs)
})

test_that("plot(type = 'fitted') does not error (regression test for ts-class dispatch bug)", {
  ## graphics::lines() must not be handed a "ts"-classed x-axis alongside a
  ## separate plain-numeric y vector: on some R versions this dispatches to
  ## stats:::lines.ts() and raises "invalid plot type" inside plot.xy().
  res <- cart_arima(.x_stat, p_set = 0:1, d_set = 0L, q_set = 0:1)
  pdf(NULL)
  on.exit(dev.off())
  expect_silent(plot(res, type = "fitted"))
})

test_that("arima_forecast plotting does not error (regression test for ts-class dispatch bug)", {
  res <- cart_arima(.x_stat, p_set = 0:1, d_set = 0L, q_set = 0:1)
  pdf(NULL)
  on.exit(dev.off())
  expect_silent(arima_forecast(res, h = 5L, plot = TRUE))
})

## --------------------------------------------------------------------------- #
context("seasonal_strength / suggest_D")

test_that("seasonal_strength returns values in [0, 1]", {
  s <- seasonal_strength(.x_seas)
  expect_true(s$seasonal_strength >= 0 && s$seasonal_strength <= 1)
  expect_true(s$trend_strength    >= 0 && s$trend_strength    <= 1)
})

test_that("seasonal_strength errors without a period for non-ts input", {
  expect_error(seasonal_strength(as.numeric(.x_seas)), "period")
})

test_that("suggest_D returns 0 or 1", {
  D <- suggest_D(.x_seas)
  expect_true(D %in% c(0L, 1L))
})

## --------------------------------------------------------------------------- #
context("arima_cv")

test_that("arima_cv returns an arimaCV object with accuracy table", {
  res <- cart_arima(.x_ar11, p_set = 0:1, d_set = 0L, q_set = 0:1)
  cv  <- arima_cv(res, h = 1L, initial = 60L)
  expect_s3_class(cv, "arimaCV")
  expect_true("RMSE" %in% names(cv$accuracy))
  expect_gt(cv$n_folds, 0L)
})

test_that("arima_cv accuracy has one row per horizon", {
  res <- cart_arima(.x_ar11, p_set = 0:1, d_set = 0L, q_set = 0:1)
  cv  <- arima_cv(res, h = 3L, initial = 60L)
  expect_equal(nrow(cv$accuracy), 3L)
})

test_that("arima_cv errors when initial leaves no room for evaluation", {
  res <- cart_arima(.x_ar11, p_set = 0:1, d_set = 0L, q_set = 0:1)
  expect_error(arima_cv(res, h = 1L, initial = 100L), "no observations")
})

test_that("print.arimaCV runs without error", {
  res <- cart_arima(.x_ar11, p_set = 0:1, d_set = 0L, q_set = 0:1)
  cv  <- arima_cv(res, h = 1L, initial = 60L)
  expect_output(print(cv), "Rolling-Origin Cross-Validation")
})