Package {maxtwin}

Title: A Robust Deterministic Data Splitting Method
Version: 0.1.1
Description: Provides a deterministic method for splitting data into training and testing sets that maximizes the distance between training and testing points while preserving their distributions. See Joseph and Cho (2026) <doi:10.1080/10618600.2026.2754089> for details. This work is supported by U.S. National Science Foundation grant DMS-2310637.
License: GPL (≥ 3)
Encoding: UTF-8
Depends: R (≥ 3.5.0)
Imports: FNN, twinning (≥ 1.1)
Suggests: testthat (≥ 3.0.0)
Config/testthat/edition: 3
RoxygenNote: 7.3.3
NeedsCompilation: no
Packaged: 2026-10-29 13:47:50 UTC; ycho397
Author: Youngseo Cho [aut, cre], V. Roshan Joseph [aut]
Maintainer: Youngseo Cho <ycho397@gatech.edu>
Repository: CRAN
Date/Publication: 2026-11-02 21:00:36 UTC

Initial MaxTwin (Step 1)

Description

Selects testing points by maximizing nearest-neighbor distance within each Voronoi cell induced by the twinning set of Vakayil and Joseph (2022). This is Step 1 of the algorithm of Joseph and Cho (2026), without iterative improvement of the separation distance.

Usage

initialmaxtwin(
  X,
  Y = NULL,
  splitratio,
  initialization = TRUE,
  standardization = TRUE
)

Arguments

X

Matrix or data frame of inputs (N x p). Columns may be numeric or factors.

Y

Optional numeric or factor vector of outputs (length N). If supplied, it is included in twinning. If omitted, splitting is based on X alone.

splitratio

Fraction of data for the testing set. Must be 1/r where r is an integer >= 2 (e.g., 0.5, 0.25, 0.2, 0.1). This constraint comes from the twinning algorithm.

initialization

Logical. If TRUE, start twinning from the point closest to the data centroid.

standardization

Logical. If TRUE, standardize the encoded data used for twinning before selection.

Value

A vector of testing point indices.

References

Vakayil, A. and Joseph, V.R. (2022). Data Twinning. Statistical Analysis and Data Mining: The ASA Data Science Journal, 15(5), 598–610. doi:10.1002/sam.11574.

Joseph, V.R. and Cho, Y. (2026). A Robust Deterministic Data Splitting Method for Model Validation. Journal of Computational and Graphical Statistics, 1–16.
doi:10.1080/10618600.2026.2754089.

MaxTwin Algorithm

Description

Implements the robust deterministic data-splitting algorithm of Joseph and Cho (2026). Step 1 selects testing points using twinning, Voronoi partitioning, and maximum nearest-neighbor distance. Step 2 iteratively improves separation if needed.

Usage

maxtwin(X, Y = NULL, splitratio, initialization = TRUE, standardization = TRUE)

Arguments

X

Matrix or data frame of inputs (N x p). Columns may be numeric or factors.

Y

Optional numeric or factor vector of outputs (length N). If supplied, it is included in twinning. If omitted, splitting is based on X alone.

splitratio

Fraction of data for the testing set. Must be 1/r where r is an integer >= 2 (e.g., 0.5, 0.25, 0.2, 0.1). This constraint comes from the twinning algorithm.

initialization

Logical. If TRUE, start twinning from the point closest to the data centroid.

standardization

Logical. If TRUE, standardize the encoded data used for twinning before selection.

Details

See Joseph and Cho (2026) for details.

Note that splitratio must be 1/r for integer r >= 2, due to the constraint of the underlying twinning algorithm.

Value

A vector of testing point indices.

References

Joseph, V.R. and Cho, Y. (2026). A Robust Deterministic Data Splitting Method for Model Validation. Journal of Computational and Graphical Statistics, 1–16.
doi:10.1080/10618600.2026.2754089.

Examples

set.seed(42)
X <- matrix(runif(200), ncol = 2)
Y <- sin(X[,1]) + cos(X[,2])
idx <- maxtwin(X, Y, splitratio = 0.2)
X_test  <- X[idx, ]
X_train <- X[-idx, ]