tests/testthat/test_imputation.R

In DrylandEcology/rSW2utils: Utility Tools for SOILWAT2 and STEPWAT2 Simulation Experiments

#---TESTS
test_that("Impute missing values", {
  #--- Prepare test data
  Nrows <- 30
  Nmiss <- 5

  df_template0 <- data.frame(
    linear = seq_len(Nrows),
    all_missing = NA,
    all_same = 1,
    cyclic = cos(2 * pi * seq_len(Nrows) / Nrows)
  )

  vars <- colnames(df_template0)
  vars_with_values <- vars[!apply(df_template0, 2, anyNA)]

  types <- c("mean", "locf", "interp")
  cyclicity <- c(FALSE, TRUE)



  for (k0 in c("dataframe", "matrix")) {
    df_template <- switch(
      EXPR = k0,
      dataframe = df_template0,
      matrix = as.matrix(df_template0)
    )

    #------- 'None' imputation returns inputs unchanged
    expect_identical(
      impute_df(df_template, imputation_type = "none"),
      df_template
    )


    #-------
    # Check 1: If every value is missing, then no numeric values are imputed.
    # Check 2: If missing values remain, then a warning is issued.
    # Check 3: If no value is missing, then no value is modified.
    df1 <- df_template

    for (type in types) {
      expect_warning(
        res1 <- impute_df( # nolint: implicit_assignment_linter.
          x = df1,
          imputation_type = type
        )
      )
      expect_identical(df1, res1)
    }


    #------- Last values are missing
    df2 <- df_template[, vars_with_values]
    idsNA <- (Nrows - Nmiss):Nrows
    df2[idsNA, ] <- NA

    for (type in types) {
      res2 <- expect_silent(
        impute_df(
          x = df2,
          imputation_type = type,
          cyclic = FALSE
        )
      )

      for (v in vars_with_values) {
        expect_false(anyNA(res2[, v]))

        if (type == "locf") {
          expect_true(all(res2[idsNA, v] == df2[idsNA[[1L]] - 1, v]))
        }
      }
    }


    #------- First values are missing
    df3 <- df_template[, vars_with_values]
    idsNA <- 1:Nmiss
    df3[idsNA, ] <- NA

    for (type in types) {
      for (cyclic in cyclicity) {

        if (type == "locf" && !cyclic) {
          expect_warning(
            res3 <- impute_df( # nolint: implicit_assignment_linter.
              x = df3,
              imputation_type = type,
              cyclic = cyclic
            )
          )

        } else {
          expect_silent(
            res3 <- impute_df( # nolint: implicit_assignment_linter.
              x = df3,
              imputation_type = type,
              cyclic = cyclic
            )
          )
        }

        for (v in vars_with_values) {
          if (type == "locf") {
            if (cyclic) {
              # last value is imputed
              expect_true(all(res3[idsNA, v] == df3[Nrows, v]))
            } else {
              # no value to can be imputed --> NAs
              expect_true(all(is.na(res3[idsNA, v])))
            }
          } else if (type == "mean") {
            expect_false(anyNA(res3[, v]))
          }
        }
      }
    }
  }
})