Run `missForest` at a Single `mtry` and Report Variablewise OOB Error

Fits [missForest::missForest()] once at a given `mtry`, returning both the completed data and a tidy, per-column out-of-bag error table. This is the unit of work swept over by [missforest_sweep_mtry()].

Usage

missforest_oob_by_mtry(data, mtry, ntree = 100, maxiter = 10)

Arguments

data: A data frame or tibble containing the columns to impute. Any identifier or time columns should already be removed; see [missforest_sweep_mtry()]'s `exclude` argument. Columns that are entirely `NA` are refused – see Details.
mtry: Number of variables randomly sampled as candidates at each split. A single positive integer.
ntree: Number of trees per forest. Passed to [missForest::missForest()]. Default `100`.
maxiter: Maximum number of imputation iterations. Passed to [missForest::missForest()]. Default `10`.

Value

A list of two elements:

`ximp`: A tibble of the completed (imputed) data.
`oob_error`: A tibble with one row per column of `data`, and columns `column`, `error_type` (`"MSE"` for numeric columns, `"PFC"` for factors), `error`, and `mtry`.

Details

`missForest` requires every column to be numeric or a factor, so character columns are converted to factors here. The conversion is idempotent – a caller that has already converted its own characters sees no change – which is what lets this single function replace call sites that previously converted at different points.

Columns that are entirely `NA` are refused with an error. [missForest::missForest()] silently drops them, which both breaks the per-column error table this function returns and would quietly shrink the predictor set that swept `mtry` values index into. Such columns arise easily – a mapping-driven reader that fills absent source columns with a bare `NA` produces full-length logical columns that pass name-based schema checks. Drop or fill them before imputing.

Reproducibility

`missForest` is stochastic. This function does not set a seed – seeding is the caller's responsibility, and [missforest_sweep_mtry()] handles it in a parallel-safe way. See that function's Reproducibility section.

Examples

# \donttest{
if (requireNamespace("missForest", quietly = TRUE)) {
  set.seed(1)
  df <- data.frame(a = c(1, 2, NA, 4, 5), b = c(2, NA, 6, 8, 10))
  res <- missforest_oob_by_mtry(df, mtry = 1)
  res$oob_error
}
#> # A tibble: 2 × 4
#>   column error_type error  mtry
#>   <chr>  <chr>      <dbl> <int>
#> 1 a      MSE         4.61     1
#> 2 b      MSE        14.6      1
# }