robCompositions (version 2.6.0)

contaminate_simplex: Generate cellwise-contaminated compositional data

Description

Introduces cellwise contamination on the simplex by replacing individual raw parts with contaminated values and re-closing the composition.

Usage

contaminate_simplex(
  x,
  epsilon = 0.05,
  delta = 10,
  type = "multiplicative",
  seed = NULL
)

Value

A list with components:

x_contaminated

the contaminated compositions (re-closed to same row sums)

flags

n x D logical matrix indicating which cells were contaminated

epsilon

contamination rate used

Arguments

x

clean compositional data (n x D matrix or data.frame with positive values)

epsilon

per-cell contamination probability (default: 0.05)

delta

contamination factor (default: 10)

type

"multiplicative" (x_j * delta), "replacement" (x_j = delta), or "additive" (x_j + delta * median(row))

seed

random seed for reproducibility

Author

Matthias Templ

Details

For each cell independently, contamination occurs with probability epsilon. Three contamination types are available: "multiplicative" multiplies the cell by delta, "replacement" sets the cell to delta, and "additive" adds delta times the row median. After contamination, affected rows are re-closed to preserve the original row sums. The "multiplicative" type, with per-cell rate epsilon and magnitude delta, matches the scale-invariant cellwise contamination model used in the simulation study of Templ (2026).

References

Templ, M. (2026). Log-ratio propagation on the simplex: a theory of cellwise contamination for compositional data. arXiv 2605.31345.

See Also

cellPcaCoDa (the estimator this model is designed to stress-test), constSum

Examples

data(arcticLake)
cont <- contaminate_simplex(arcticLake, epsilon = 0.1, delta = 10, seed = 42)
sum(cont$flags)  # number of contaminated cells
rowSums(cont$x_contaminated)  # should match rowSums(arcticLake)