Detect Whether an Object Looks Like a Confusion/Coincidence Matrix

Checks if x behaves like a rater-by-rater contingency table: square 2D numeric (integer-like) counts (or, optionally, proportions), non-negative, finite, and (optionally) with matching row/column names.

Usage

isConfusionTable(
  x,
  requireDimnames = TRUE,
  requireSameLevels = TRUE,
  integerTol = sqrt(.Machine$double.eps),
  acceptProportions = TRUE,
  requireSquare = TRUE
)

Arguments

x

object to check, typically a table, matrix, or numeric data.frame

requireDimnames

logical; if TRUE, both row and column names must be present. Defaults to TRUE.

requireSameLevels

logical; if TRUE and dimnames are present, row and column names must be the same set (order ignored). Defaults to TRUE.

integerTol

numeric tolerance for integer-like counts; defaults to sqrt(.Machine$double.eps)

acceptProportions

logical; if TRUE, proportion tables are accepted when all entries are in $[0, 1]$ and their sum is approximately

Defaults to TRUE.

requireSquare

logical; whether to require a square table; defaults to TRUE

Value

TRUE if x looks like a confusion or coincidence matrix, otherwise FALSE

Examples

tab <- table(sample(letters[1:3], 100, TRUE),
             sample(letters[1:3], 100, TRUE))
isConfusionTable(tab)               # TRUE
#> [1] TRUE

M <- as.matrix(tab)
isConfusionTable(M)                 # TRUE (dimnames present)
#> [1] TRUE
isConfusionTable(unname(M), requireDimnames = FALSE)  # TRUE without names
#> [1] TRUE

df <- as.data.frame.matrix(tab)
isConfusionTable(df)                # TRUE (numeric data.frame)
#> [1] TRUE

# Two-column raw ratings are NOT a confusion table:
ratings <- cbind(r1 = sample(0:1, 50, TRUE), r2 = sample(0:1, 50, TRUE))
isConfusionTable(ratings)           # FALSE (not square)
#> [1] FALSE