tests/testthat/test-eval_quantile_from_cdf.R

In vincenzocoia/distionary: Create and Evaluate Univariate Probability Distributions

test_that("Quantile function algorithm works for several distributions", {
  p <- 1:99/100
  d <- list(
    dst_beta(3, 5),
    dst_binom(15, 0.2),
    dst_norm(0, 1),
    dst_norm(3.87e287, 1),
    dst_norm(3.87e287, 1e-200),
    dst_nbinom(10, 0.5),
    dst_exp(10),
    dst_gpd(-100, 342, 3)
  )
  for (d_ in d) {
    expect_equal(
      distionary:::eval_quantile.dst(d_, at = p),
      eval_quantile(d_, at = p),
      tolerance = 1e-6
    )
  }
})

test_that("Quantile algorithm handles NA appropriately.", {
  pfoo <- pnorm
  d <- dst_parametric("foo", .variable = "continuous")
  expect_equal(eval_quantile(d, at = NA), NA_real_)
  expect_equal(eval_quantile(d, at = NaN), NaN)
  expect_length(eval_quantile(d, at = numeric(0L)), 0)
  expect_equal(
    eval_quantile(d, at = c(0.3, NA, NaN)),
    c(stats::qnorm(0.3), NA, NaN)
  )
  psteep <- function(x) pmax(pmin(10 * (x - 30), 1), 0)
  d_steep <- dst_parametric("steep", .variable = "continuous")
  expect_equal(psteep(eval_quantile(d_steep, at = 0:10 / 10)), 0:10 / 10)
  pfixed <- function(x) as.numeric(x >= -1e30)
  d_fixed <- dst_parametric("fixed", .variable = "continuous")
  expect_equal(eval_quantile(d_fixed, at = 0:10 / 10), rep(-1e30, 11))
  pflat <- function(x) {
    res <- rep(0.5, length(x))
    res[x < 0.5] <- pmax(x[x < 0.5], 0)
    res[x > 100] <- pmin(x[x > 100] - 99.5, 1)
    res
  }
  d_flat <- dst_parametric("flat", .variable = "continuous")
  expect_equal(eval_quantile(d_flat, at = 0.5), 0.5)
})