tests/testthat/test-pairwise-cor.R

In widyr: Widen, Process, then Re-Tidy Data

context("pairwise_cor")

suppressPackageStartupMessages(library(dplyr))

d <- tibble(col = rep(c("a", "b", "c"), each = 3),
            row = rep(c("d", "e", "f"), 3),
            value = c(1, 2, 3, 6, 5, 4, 7, 9, 8))

test_that("pairwise_cor computes pairwise correlations", {
  ret <- d %>%
    pairwise_cor(col, row, value)

  ret1 <- ret$correlation[ret$item1 == "a" & ret$item2 == "b"]
  expect_equal(ret1, cor(1:3, 6:4))

  ret2 <- ret$correlation[ret$item1 == "b" & ret$item2 == "c"]
  expect_equal(ret2, cor(6:4, c(7, 9, 8)))

  expect_equal(sum(ret$item1 == ret$item2), 0)
})

test_that("pairwise_cor can compute Spearman correlations", {
  ret <- d %>%
    pairwise_cor(col, row, value, method = "spearman")

  ret1 <- ret$correlation[ret$item1 == "a" & ret$item2 == "b"]
  expect_equal(ret1, -1)
})

test_that("pairwise_cor works on binary matrices", {
  cors <- data.frame(x = c("a", "a", "a", "b", "b", "b", "c", "c", "c"),
                     y = c(1, 2, 3, 1, 2, 3, 1, 3, 4)) %>%
    pairwise_cor(x, y, sort = TRUE)

  expect_equal(colnames(cors), c("item1", "item2", "correlation"))
  expect_equal(cors$correlation, rep(c(1, - 1 / 3), c(2, 4)))
})

test_that("pairwise_cor retains factor levels", {
  d$col <- factor(d$col, levels = c("b", "c", "a"))

  ret <- d %>%
    pairwise_cor(col, row, value, method = "spearman")

  expect_is(ret$item1, "factor")
  expect_is(ret$item2, "factor")
  expect_equal(levels(ret$item1), c("b", "c", "a"))
  expect_equal(levels(ret$item2), c("b", "c", "a"))
})