tests/testthat/test-impute_LS_combined.R

In missMethods: Methods for Missing Data

## Basic tests for impute_LS_combined() -------------------------------------------
test_that("impute_LS_combined() works (basic test, only check for anyNA)", {
  set.seed(1234)
  ds_mis <- mvtnorm::rmvnorm(20, rep(0, 5), diag(1, 5))
  ds_mis <- delete_MCAR(ds_mis, 0.2, 1:4)
  expect_false(anyNA(impute_LS_combined(ds_mis)))
})

test_that("impute_LS_combined() works for small data frames", {
  set.seed(123)
  expect_false(anyNA(impute_LS_combined(data.frame(X = 1:11, Y = c(1:10, NA)))))
})

## Test verbosity -------------------------------------------------------------
test_that("impute_LS_combined() works with completely missing row and verbose", {
  set.seed(1234)
  ds_mis <- mvtnorm::rmvnorm(20, rep(0, 7), diag(1, 7))
  ds_mis[5, ] <- NA

  # silent
  ds_imp_silent <- expect_silent(
    impute_LS_combined(ds_mis,
      verbose_gene = FALSE, verbose_array = FALSE,
      verbose_gene_p = FALSE, verbose_array_p = FALSE
    )
  )
  expect_false(anyNA(ds_imp_silent))
  # Completely missing rows are imputed with observed column means from LS_gene
  # and LS_array. So, independent of p, the imputed values will be these means.
  expect_equal(ds_imp_silent[5, ], suppressWarnings(colMeans(impute_LS_gene(ds_mis))))

  # verbose_gene
  ds_imp_verb1 <- expect_message(
    impute_LS_combined(ds_mis, verbose_gene = TRUE, verbose_array = FALSE),
    "No observed value in row(s) 5. These rows were imputed with column means.",
    fixed = TRUE,
    all = TRUE
  )
  expect_equal(ds_imp_verb1, ds_imp_silent)

  # verbose_array
  ds_imp_verb2 <- expect_message(
    impute_LS_combined(ds_mis, verbose_gene = FALSE, verbose_array = TRUE),
    "The missing values of following rows were imputed with (parts of) mu: 5",
    fixed = TRUE,
    all = TRUE
  )
  expect_equal(ds_imp_verb2, ds_imp_silent)

  # verbose_gene and verbose_array
  verify_output(
    test_path("test-impute_LS_combined-verbosity.txt"),
    ds_imp_verb3 <- impute_LS_combined(ds_mis, verbose_gene = TRUE, verbose_array = TRUE)
  )
  expect_equal(ds_imp_verb3, ds_imp_silent)

  # verbose_gene_p
  ds_imp_verb4 <- expect_message(
    impute_LS_combined(ds_mis, verbose_gene_p = TRUE, verbose_array_p = FALSE),
    "No observed value in row(s) 5. These rows were imputed with column means.",
    fixed = TRUE,
    all = TRUE
  )
  expect_equal(ds_imp_verb4, ds_imp_silent)

  # verbose_array_p
  ds_imp_verb5 <- expect_message(
    impute_LS_combined(ds_mis, verbose_gene_p = FALSE, verbose_array_p = TRUE),
    "The missing values of following rows were imputed with (parts of) mu: 5",
    fixed = TRUE,
    all = TRUE
  )
  expect_equal(ds_imp_verb5, ds_imp_silent)
})


## Comparing impute_LS_combined() to original results from Bo et al. ----------
# For some remarks see test-impute_LS_gene.R

test_that("impute_LS_combined() works with dataset triangle miss", {
  # The missing values in this file were created with upper.tri, which results in a monotone pattern.
  # The rows 1:15 have 1:15 observed values.
  ds_triangle_mis <- readRDS(test_path(file.path("datasets", "ds_triangle_mis.rds")))
  ds_triangle_LS_array_Bo <- readRDS(test_path(file.path("datasets", "ds_triangle_LS_array_Bo.rds")))
  ds_triangle_LS_gene_Bo <- readRDS(test_path(file.path("datasets", "ds_triangle_LS_gene_Bo.rds")))

  set.seed(1234)
  ds_imp <- round(impute_LS_combined(ds_triangle_mis, min_common_obs = 5), 3)
  # Check that LS_combined is a mixture of LS_array and LS_gene:
  expect_true(all(ds_imp <= pmax(ds_triangle_LS_array_Bo, ds_triangle_LS_gene_Bo)))
  expect_true(all(ds_imp >= pmin(ds_triangle_LS_array_Bo, ds_triangle_LS_gene_Bo)))

  ds_triangle_LS_combined_Bo <- readRDS(test_path(file.path("datasets", "ds_triangle_LS_combined_Bo.rds")))

  # There is some randomness in impute_LS_combined, that we cannot control in the jar-file
  # So we need some tolerance, because exact matches would be rather surprising
  expect_equal(
    ds_triangle_LS_array_Bo,
    ds_imp,
    tolerance = 0.5
  )
})


test_that("impute_LS_combined() imputes like Bo et al. (2004) (MCAR, 100x7)", {
  ds_100x7_LS_array_Bo <- readRDS(test_path(file.path("datasets", "ds_100x7_LS_array_Bo.rds")))
  ds_100x7_LS_gene_Bo <- readRDS(test_path(file.path("datasets", "ds_100x7_LS_gene_Bo.rds")))

  ds_100x7_mis_MCAR <- readRDS(test_path(file.path("datasets", "ds_100x7_mis_MCAR.rds")))
  # Cure some rounding problems due to saving:
  ds_mis <- ds_100x7_LS_gene_Bo
  ds_mis[is.na(ds_100x7_mis_MCAR)] <- NA

  set.seed(1234)
  ds_imp <- round(impute_LS_combined(ds_mis, min_common_obs = 5), 3)

  # Check that LS_combined is a mixture of LS_array and LS_gene:
  expect_true(all(ds_imp <= pmax(ds_100x7_LS_array_Bo, ds_100x7_LS_gene_Bo)))
  expect_true(all(ds_imp >= pmin(ds_100x7_LS_array_Bo, ds_100x7_LS_gene_Bo)))

  ds_100x7_LS_combined_Bo <- readRDS(test_path(file.path("datasets", "ds_100x7_LS_combined_Bo.rds")))

  # Need some tolerance because of randomness:
  expect_equal(ds_100x7_LS_combined_Bo, ds_imp, tolerance = 0.1)

  # Conclusion: Both methods seem to return the "same" imputation values (deviations because of randomness)
})