tests/testthat/test_roc_pivot.R

In ModelOriented/FairModels: Flexible Tool for Bias Detection, Visualization, and Mitigation

test_that("test ROC pivot", {
  df <- data.frame(
    sex = as.factor(c(rep("M", 5), rep("F", 5), rep("N", 5))),
    target = c(1, 1, 1, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 1),
    name = as.character(1:15),
    probs = c(0.9, 0.82, 0.54, 0.78, 0.455, 0.12, 0.48, 0.63, 0.48, 0.88, 0.34, 0.12, 0.34, 0.49, 0.9)
  )

  lr_model <- glm(target ~ sex + name, data = df, family = binomial())

  # explainer will have no y_hats close to border - no difference
  explainer <- suppressWarnings(DALEX::explain(lr_model, data = df[, c("sex", "name")], y = df$target))
  explainer2 <- roc_pivot(explainer, protected = df$sex, "M")

  expect_equal(explainer$y_hat, explainer2$y_hat)

  explainer$y_hat <- df$probs
  explainer2 <- roc_pivot(explainer, protected = df$sex, "M", theta = 0.05)

  # changed probs
  expect_equal(explainer2$y_hat, c(0.9, 0.82, 0.46, 0.78, 0.455, 0.12, 0.52, 0.63, 0.52, 0.88, 0.34, 0.12, 0.34, 0.51, 0.9))

  # errors
  explainer2$model_info$type <- "not_classification"
  expect_error(roc_pivot(explainer2, protected = df$sex, "M", theta = 0.05))
  expect_output(roc_pivot(explainer, protected = as.character(df$sex), "M", theta = 0.05))
  expect_error(roc_pivot(explainer, protected = df$sex, "M", theta = 0.05, cutoff = 2))
  expect_error(roc_pivot(explainer, protected = df$sex, "M", theta = 0.05, cutoff = c(0.1, 0.5)))
  suppressWarnings(expect_error(roc_pivot(explainer, protected = df$sex, "M", theta = 0.05, cutoff = 0.5, theta = 3)))
  expect_error(roc_pivot(explainer, protected = df$sex, "M", theta = 0.05, cutoff = 0.5, theta = c(0.3, 0.1)))
})