skip_on_cran()
skip_if_not_installed("SVEMnet")

test_that("svem_random_table_multi validates schemas and predicts", {
  set.seed(7)
  n <- 40
  X1 <- runif(n); X2 <- runif(n)
  A <- runif(n); B <- runif(n); C <- pmax(0, 1 - A - B)
  F <- factor(sample(c("lo","hi"), n, TRUE))
  y <- 1 + 2*X1 - X2 + 3*A + 1.5*B + 0.5*C + (F=="hi") + rnorm(n, 0, 0.3)
  z <- 0.5 + 0.8*X1 + 0.4*X2 + rnorm(n, 0, 0.2)
  d <- data.frame(y, z, X1, X2, A, B, C, F)

  f1 <- SVEMnet::SVEMnet(y ~ X1 + X2 + A + B + C + F, d, nBoot = 30, glmnet_alpha = 1)
  f2 <- SVEMnet::SVEMnet(z ~ X1 + X2 + A + B + C + F, d, nBoot = 30, glmnet_alpha = 1)

  res <- SVEMnet::svem_random_table_multi(list(f1, f2), n = 50)

  expect_true(is.list(res))
  expect_true(all(c("data","pred","all") %in% names(res)))
  expect_true(all(c("X1","X2","A","B","C","F") %in% names(res$data)))

  # --- UPDATED: allow old names (y/z or y.pred/z.pred) OR new *_pred names ---
  pred_names <- names(res$pred)
  has_old_plain <- all(c("y", "z") %in% pred_names)
  has_old_dot   <- all(c("y.pred", "z.pred") %in% pred_names)
  has_new_suf   <- all(c("y_pred", "z_pred") %in% pred_names)

  expect_true(has_old_plain || has_old_dot || has_new_suf)

  expect_equal(nrow(res$data), 50)
  expect_equal(nrow(res$pred), 50)
  expect_equal(nrow(res$all), 50)
})



test_that("svem_random_table_multi respects blocking for numeric and categorical", {
  set.seed(2)
  df_block <- data.frame(
    y1         = rnorm(40),
    y2         = rnorm(40),
    X1         = runif(40),
    X2         = runif(40),
    Operator   = factor(sample(paste0("Op", 1:3), 40, TRUE)),
    AmbientTmp = rnorm(40, mean = 22, sd = 2)
  )

  spec_block <- SVEMnet::bigexp_terms(
    y1 ~ X1 + X2,
    data             = df_block,
    factorial_order  = 2,
    polynomial_order = 2,
    blocking         = c("Operator", "AmbientTmp")
  )

  fit_b1 <- SVEMnet::SVEMnet(
    spec_block, df_block,
    response = "y1",
    nBoot    = 20,
    glmnet_alpha = 1
  )
  fit_b2 <- SVEMnet::SVEMnet(
    spec_block, df_block,
    response = "y2",
    nBoot    = 20,
    glmnet_alpha = 1
  )

  res <- SVEMnet::svem_random_table_multi(list(fit_b1, fit_b2), n = 200)

  expect_true(is.list(res))
  expect_true(all(c("data","pred","all") %in% names(res)))

  ## Blocking variables are present
  expect_true(all(c("Operator", "AmbientTmp") %in% names(res$data)))

  ## Operator (factor) is held fixed at a single level
  expect_equal(length(unique(res$data$Operator)), 1L)

  ## AmbientTmp (numeric) is held fixed at a single value (up to tiny numerical noise)
  expect_true(diff(range(res$data$AmbientTmp)) < 1e-8)

  ## Row counts align
  expect_equal(nrow(res$data), 200)
  expect_equal(nrow(res$pred), 200)
  expect_equal(nrow(res$all), 200)

  ## Prediction column names: allow old and new conventions
  pred_names <- names(res$pred)
  has_old_plain <- all(c("y1", "y2") %in% pred_names)
  has_old_dot   <- all(c("y1.pred", "y2.pred") %in% pred_names)
  has_new_suf   <- all(c("y1_pred", "y2_pred") %in% pred_names)

  expect_true(has_old_plain || has_old_dot || has_new_suf)
})