tests/testthat/test_pipeop_smote.R

In mlr3pipelines: Preprocessing Operators and Pipelines for 'mlr3'

context("PipeOpSmote")

test_that("PipeOpSmote - basic properties", {
  skip_if_not_installed("smotefamily")
  set.seed(1234)
  data = smotefamily::sample_generator(1000, ratio = 0.80)
  data$result = as.factor(data$result)
  task = TaskClassif$new(id = "unbalanced", backend = data, target = "result")

  expect_datapreproc_pipeop_class(PipeOpSmote,
    task = task, predict_like_train = FALSE, deterministic_train = FALSE)

  op = PipeOpSmote$new(param_vals = list(K = 3))
  set.seed(1234)
  result = train_pipeop(op, inputs = list(task))

  expect_pipeop(op)
  expect_task(result[[1]])
})

test_that("compare to smotefamily::SMOTE", {
  skip_if_not_installed("smotefamily")
  set.seed(1234)
  data = smotefamily::sample_generator(1000, ratio = 0.80)
  data$result = as.factor(data$result)
  task = TaskClassif$new(id = "unbalanced", backend = data, target = "result")
  op = PipeOpSmote$new(param_vals = list(K = 3))
  set.seed(1234)
  result = train_pipeop(op, inputs = list(task))

  set.seed(1234)
  st = invoke(smotefamily::SMOTE, X = data[, -3], target = data[, 3], K = 3, .opts = list(warnPartialMatchArgs = FALSE))
  expect_equal(result[[1]]$data()[1001:nrow(st$data), c(2:3)], as.data.table(st$syn_data)[, 1:2])
  expect_equal(as.character(result[[1]]$data()[1001:nrow(st$data), result]), st$syn_data[, 3])
})

test_that("PipeOpSmote - handling of feature named 'class'", {
  skip_if_not_installed("smotefamily")

  op = PipeOpSmote$new()

  df = data.frame(
    target = factor(sample(c("c1", "c2"), size = 200, replace = TRUE, prob = c(0.1, 0.9))),
    class = rnorm(200),
    x = rnorm(200)
  )
  task = TaskClassif$new(id = "test", backend = df, target = "target")

  set.seed(1234L)
  train_out = op$train(list(task))[[1]]$data()
  set.seed(1234L)
  df_out = invoke(smotefamily::SMOTE, X = task$data(cols = task$feature_names), target = task$truth(),
                  .opts = list(warnPartialMatchArgs = FALSE))$syn_data
  # Renaming by position
  setnames(df_out, 3, "target")
  smote_out = setDT(rbind(df, df_out))

  expect_equal(train_out, smote_out)

})