tests/testthat/test-performance.R

In jeanimal/heuristica: Heuristics Including Take the Best and Unit-Weight Linear

context("performance")

# require('testthat')

# confusionMatrixRequiredCategories

test_that("confusionMatrixRequiredCategories match all factors in data", {
  data1 <- c(0,1,1)
  data2 <- c(0,1,1)
  out <- confusionMatrixRequiredCategories(data1, data2, c(0,1))
  expected <- cbind(c(1,0), c(0,2))
  expect_equal(expected, matrix(out, 2, 2))
  expect_equal(c("0", "1"), dimnames(out)[[1]])
  expect_equal(c("0", "1"), dimnames(out)[[2]])
})

test_that("confusionMatrixRequiredCategories non-match all factors in data", {
  data1 <- c(0,0,1,1)
  data2 <- c(0,1,0,1)
  out <- confusionMatrixRequiredCategories(data1, data2, c(0,1))
  expected <- cbind(c(1,1), c(1,1))
  expect_equal(expected, matrix(out, 2, 2))
  expect_equal(c("0", "1"), dimnames(out)[[1]])
  expect_equal(c("0", "1"), dimnames(out)[[2]])
})

test_that("confusionMatrixRequiredCategories non-match missing -1 factor in data", {
  data1 <- c(0,0,1,1)
  data2 <- c(0,1,0,1)
  out <- confusionMatrixRequiredCategories(data1, data2, c(-1,0,1))
  expected <- cbind(c(0,0,0), c(0,1,1), c(0,1,1))
  expect_equal(expected, matrix(out, 3, 3))
  expect_equal(c("-1", "0", "1"), dimnames(out)[[1]])
  expect_equal(c("-1", "0", "1"), dimnames(out)[[2]])
})

test_that("confusionMatrixRequiredCategories non-match missing 2 factor in data", {
  data1 <- c(0,0,1,1)
  data2 <- c(0,1,0,1)
  out <- confusionMatrixRequiredCategories(data1, data2, c(0,1,2))
  expected <- cbind(c(1,1,0), c(1,1,0), c(0,0,0))
  expect_equal(expected, matrix(out, 3, 3))
  expect_equal(c("0", "1", "2"), dimnames(out)[[1]])
  expect_equal(c("0", "1", "2"), dimnames(out)[[2]])
})

test_that("confusionMatrixRequiredCategories missing 1 factor in data1", {
  data1 <- c(0,0,0)
  data2 <- c(0,1,1)
  out <- confusionMatrixRequiredCategories(data1, data2, c(0,1))
  expected <- cbind(c(1,0), c(2,0))
  expect_equal(expected, matrix(out, 2, 2))
  expect_equal(c("0", "1"), dimnames(out)[[1]])
  expect_equal(c("0", "1"), dimnames(out)[[2]])
})

test_that("confusionMatrixRequiredCategories missing 1 factor in data2", {
  data1 <- c(0,1,1)
  data2 <- c(0,0,0)
  out <- confusionMatrixRequiredCategories(data1, data2, c(0,1))
  expected <- cbind(c(1,2), c(0,0))
  expect_equal(expected, matrix(out, 2, 2))
  expect_equal(c("0", "1"), dimnames(out)[[1]])
  expect_equal(c("0", "1"), dimnames(out)[[2]])
})


# confusionMatrixFor_Neg1_0_1
# (Most tests of this are via the more generic (but non-public)
# confusionMatrix function.)

test_that("confusionMatrixFor_Neg1_0_1 missing -1 factor in data", {
  # Below, the correct outcome is always 1, so only the last row of the
  # confusion matrix has non-zero counts.  But the predictor makes a few
  # mistakes, so some non-zero counts are off the diagonal.
  out <- confusionMatrixFor_Neg1_0_1(c(1,1,1), c(1,0,0))
  expected <- cbind(c(0,0,0), c(0,0,2), c(0,0,1))
  expect_equal(expected, matrix(out, 3, 3))
  expect_equal(c("-1", "0", "1"), dimnames(out)[[1]])
  expect_equal(c("-1", "0", "1"), dimnames(out)[[2]])
})

# accuracyFromConfusionMatrix3x3

test_that("accuracyFromConfusionMatrix3x3 1", {
  # Below accuracy is 1 (100% correct) because 4 -1's were correctly predicted,
  # and 2 1's were correctly predicted.  (On-diagonal elements are correct
   # predictions.)
  expect_equal(1, accuracyFromConfusionMatrix3x3(
    cbind("-1"=c(4,0,0), "0"=c(0,0,0), "1"=c(0,0,2))))
})

test_that("accuracyFromConfusionMatrix3x3 0", {
  # 3 wrong and 3 more wrong for 0 accuracy.
  expect_equal(0, accuracyFromConfusionMatrix3x3(
    cbind(c(0,0,3), c(0,0,0), c(3,0,0))))
})

test_that("accuracyFromConfusionMatrix3x3 0.9", {
  # Below is 4 + 5 correct, 1 incorrect, for 9/10 = 0.9 accuracy.
  expect_equal(0.9, accuracyFromConfusionMatrix3x3(
    cbind(c(4,0,1), c(0,0,0), c(0,0,5))))
})

test_that("accuracyFromConfusionMatrix3x3 guess 0.5", {
  # Below has 3+1=4 guesses, and 0.5 are assigned correct.
  expect_equal(0.5, accuracyFromConfusionMatrix3x3(
    cbind(c(0,0,0), c(3,0,1), c(0,0,0))))
})

# TODO(jeanw): What about this?  The 2 is on-diagonal, and that guess
# is treated as automatically correct.  Should it be?
# > accuracyFromConfusionMatrix3x3(cbind(c(0,0,0), c(3,2,1), c(0,0,0)))
# [1] 0.6666667

test_that("distributeGuessAsExpectedValue", {
  m <- rbind(c("-1"=2, "0"=2, "1"=2), c("-1"=4, "0"=4, "1"=4),
             c("-1"=6, "0"=6, "1"=6))
  rownames(m) <- c("-1", "0", "1")
  out <- distributeGuessAsExpectedValue(m)
  # In cases when correct value was -1, it guessed.
  expect_equal(c("-1"=3, "0"=0, "1"=3), out["-1",])
  # There were no cases when correct value was 0.
  expect_equal(c("-1"=6, "0"=0, "1"=6), out["0",])
  # In cases when correct value was 1, it guessed.
  expect_equal(c("-1"=9, "0"=0, "1"=9), out["1",])
  
  # Check dimensions-- now there are 3.
  expect_equal(3, nrow(out))
  expect_equal(3, ncol(out))
})

test_that("distributeTies", {
  m <- cbind(c("-1"=2, "0"=2, "1"=2), c("-1"=4, "0"=4, "1"=4),
             c("-1"=6, "0"=6, "1"=6))
  colnames(m) <- c("-1", "0", "1")
  out <- distributeTies(m)
  # In cases when correct value was -1, it guessed.
  expect_equal(c("-1"=3, "0"=6, "1"=9), out["-1",])
  # There were no cases when correct value was 0.
  expect_equal(c("-1"=0, "0"=0, "1"=0), out["0",])
  # In cases when correct value was 1, it guessed.
  expect_equal(c("-1"=3, "0"=6, "1"=9), out["1",])
  
  # Check dimensions-- now there are 3.
  expect_equal(3, nrow(out))
  expect_equal(3, ncol(out))
})

test_that("collapseConfusionMatrix3x3To2x2", {
  m <- cbind(c("-1"=1, "0"=2, "1"=4), c("-1"=10, "0"=20, "1"=30),
             c("-1"=100, "0"=150, "1"=1000))
  colnames(m) <- c("-1", "0", "1")
  out <- collapseConfusionMatrix3x3To2x2(m)

  expect_equal(c("-1"=12, "1"=185), out["-1",])
  expect_equal(c("-1"=25, "1"=1095), out["1",])
  
  # Check dimensions-- now there are 2.
  expect_equal(2, nrow(out))
  expect_equal(2, ncol(out))
})

test_that("statsFromConfusionMatrix simple but asymmetric", {
  matrix <- cbind("-1"=c(2, 1), "1"=c(1,1))
  rownames(matrix) <- c("-1", "1")
  #          predictions
  #  correct "-1" "1"
  #     "-1"   2   1
  #      "1"   1   1
  stats <- statsFromConfusionMatrix(matrix)
  expect_equal(0.6, stats$accuracy)
  expect_equal(0.5, stats$sensitivity)
  expect_equal(0.6667, stats$specificity, tolerance=0.0001)
  expect_equal(0.5, stats$precision)
  # To compare with caret's confusion matrix function:
  # d <- c(rep(c("-1", "-1"), 2), rep(c("-1", "1"), 1), rep(c("1", "-1"), 1),
  #   rep(c("1", "1"), 1))
  # data_matrix <- t(matrix(d, 2, length(d)/2))
  # confusionMatrix(data_matrix[,1], data_matrix[,2], positive="1")
})

test_that("statsFromConfusionMatrix unique numbers", {
  matrix <- cbind("-1"=c(4, 2), "1"=c(3,1))
  rownames(matrix) <- c("-1", "1")
  #          predictions
  #  correct "-1" "1"
  #     "-1"   4   3
  #      "1"   2   1
  stats <- statsFromConfusionMatrix(matrix)
  expect_equal(0.5, stats$accuracy)
  expect_equal(0.25, stats$sensitivity)
  expect_equal(0.6667, stats$specificity, tolerance=0.0001)
  expect_equal(0.3333, stats$precision, tolerance=0.0001)
  # To compare with caret's confusion matrix function:
  # d <- c(rep(c("-1", "-1"), 4), rep(c("-1", "1"), 3), rep(c("1", "-1"), 2),
  #   rep(c("1", "1"), 1))
  # data_matrix <- t(matrix(d, 2, length(d)/2))
  # confusionMatrix(data_matrix[,1], data_matrix[,2], positive="1")
})