s4pca_conformal_select: Repeated split conformal selection for S4PCA

In msma: Multiblock Sparse Multivariable Analysis

Repeated split conformal selection for S4PCA

Description

Evaluates combinations of sparsity, overlap penalty, and component count by reconstruction error on repeated calibration splits. The selected candidate minimizes the median conformal reconstruction quantile subject to an optional overlap constraint.

Usage

s4pca_conformal_select(
  X,
  Z = NULL,
  lambdaX = c(0.05, 0.1, 0.2),
  gammaX = c(0, 0.05, 0.1),
  comp = 2:4,
  alpha = 0.1,
  repeats = 5,
  calibration_fraction = 0.25,
  max_overlap = 1,
  muX = 1,
  history.method = "sumabs",
  niterS4 = 30,
  scaling = TRUE,
  intseed = 1,
  refit = TRUE
)

Arguments

X	numeric data matrix.
Z	optional numeric supervision vector or one-column matrix.
lambdaX	candidate sparsity parameters.
gammaX	candidate soft-structure parameters.
comp	candidate component counts.
alpha	miscoverage level in (0,1).
repeats	positive number of repeated splits.
calibration_fraction	fraction assigned to calibration in each split.
max_overlap	maximum admissible overlap_selected.
muX	supervision weight in [0,1].
history.method	one of "sumabs", "any", or "count".
niterS4	fixed S4PCA updates per component.
scaling	logical data scaling indicator.
intseed	random seed.
refit	logical; refit the selected model on all observations.

Value

an object of class s4pca_selection.

msma documentation built on Oct. 3, 2026, 9:07 a.m.