tests/testthat/test-npksum.R

In np: Nonparametric Kernel Smoothing Methods for Mixed Data Types

test_that("npksum numeric and formula interfaces agree", {
  set.seed(1)
  n <- 30
  x <- runif(n)
  y <- rnorm(n)
  dat <- data.frame(y = y, x = x)

  k1 <- npksum(txdat = x, tydat = y, bws = 0.5)
  k2 <- npksum(y ~ x, data = dat, bws = 0.5)

  expect_s3_class(k1, "npkernelsum")
  expect_s3_class(k2, "npkernelsum")
  expect_identical(k1$ksum, k2$ksum)
})

test_that("npksum preserves 1D exdat column naming", {
  set.seed(1)
  n <- 10
  x <- runif(n)
  ex <- seq(0, 1, length.out = 3)

  k <- npksum(txdat = x, exdat = ex, bws = 0.5)
  expect_identical(colnames(k$eval), "exdat")
})

test_that("npksum positional bws dispatch matches named bws", {
  set.seed(42)
  n <- 20
  x <- runif(n)
  y <- rnorm(n)

  k_named <- npksum(txdat = x, tydat = y, bws = 0.4)
  k_pos <- npksum(0.4, txdat = x, tydat = y)

  expect_s3_class(k_pos, "npkernelsum")
  expect_identical(k_named$ksum, k_pos$ksum)
})

test_that("npksum formula subset and na.action match explicit data path", {
  set.seed(7)
  n <- 24
  dat <- data.frame(x = runif(n), y = rnorm(n))
  dat$y[c(3, 12)] <- NA_real_

  k_formula <- npksum(y ~ x, data = dat, subset = x > 0.2, na.action = na.omit, bws = 0.45)

  dat2 <- na.omit(subset(dat, x > 0.2))
  k_explicit <- npksum(txdat = dat2$x, tydat = dat2$y, bws = 0.45)

  expect_identical(k_formula$ksum, k_explicit$ksum)
})

test_that("npksum bounded nonfixed helper path works for nearest-neighbor bandwidths", {
  set.seed(20260325)
  n <- 24
  x <- runif(n)
  y <- rnorm(n)
  w <- cbind(y, 1.0)

  k_gnn <- npksum(
    txdat = x,
    tydat = w,
    weights = w,
    leave.one.out = TRUE,
    bandwidth.divide = TRUE,
    bws = 6,
    bwtype = "generalized_nn",
    ckerbound = "range"
  )
  k_adapt <- npksum(
    txdat = x,
    tydat = w,
    weights = w,
    leave.one.out = TRUE,
    bandwidth.divide = TRUE,
    bws = 7,
    bwtype = "adaptive_nn",
    ckerbound = "range"
  )

  expect_s3_class(k_gnn, "npkernelsum")
  expect_s3_class(k_adapt, "npkernelsum")
  expect_true(all(is.finite(as.numeric(k_gnn$ksum))))
  expect_true(all(is.finite(as.numeric(k_adapt$ksum))))
})