Fit dual high-dimensional RISDR — fit_risdr

Constructs a low-rank PCA score representation, fits the unified RISDR core in that score space, maps the dense and selected sparse SDR directions back to the original predictor coordinates, and ranks variables using the Euclidean loading norm across only the selected sparse directions.

Usage

fit_risdr_dual(
  X,
  y,
  delta = NULL,
  response_type = c("continuous", "categorical", "survival"),
  dual_rank = 100L,
  variance_explained = NULL,
  d = NULL,
  d_max = 10L,
  nslices = 4L,
  sdr_method = c("dr", "sir", "save", "phd"),
  cov_method = c("mec", "oas", "lw", "sample", "ridge"),
  selector = c("cicomp", "icomp", "aic", "caic", "bic"),
  lambda_grid = seq(0, 0.3, length.out = 10L),
  adapt_weights = TRUE,
  weight_mode = c("c1f", "inverse_c1f", "uniform"),
  support_penalty = FALSE,
  standardize_dual = TRUE,
  gene_select_top = NULL,
  gene_select_quantile = 0.99,
  center = TRUE,
  scale_X = TRUE,
  verbose = TRUE,
  ...
)

Arguments

X: Numeric high-dimensional predictor matrix.
y: Response vector. For survival data, observed time.
delta: Optional 0/1 event indicator.
response_type: One of "continuous", "categorical", or "survival".
dual_rank: Maximum retained PCA rank.
variance_explained: Optional cumulative variance threshold.
d: Optional fixed structural dimension. If NULL, selected by selector.
d_max: Maximum candidate structural dimension.
nslices: Number of slices, or slices per censoring-status group.
sdr_method: SDR method.
cov_method: Covariance method.
selector: Information criterion used to select (d, lambda).
lambda_grid: Sparsity tuning grid.
adapt_weights: Logical. Use adaptive penalty weights.
weight_mode: Adaptive weighting mode.
support_penalty: Logical. Include active support in criterion df.
standardize_dual: Logical. Standardise retained PCA scores inside fit_risdr_sparse().
gene_select_top: Optional number of top-ranked variables.
gene_select_quantile: Quantile used when gene_select_top is NULL.
center: Logical. Centre predictors before dual PCA.
scale_X: Logical. Scale predictors before dual PCA.
verbose: Logical. Emit progress messages.
...: Additional arguments passed to fit_risdr_sparse().

Value

An object of class risdr_dual.

Details

The inner RISDR standardisation scale is explicitly undone before mapping directions through the PCA loadings. The returned reduced predictors are taken directly from the fitted core and are therefore exactly aligned with the model used for information-criterion selection.

This entry point uses the frozen sparse thesis engine. The existing fit_risdr() and fit_risdr_realdata() interfaces retain their continuous-response behaviour from version 0.3.1. Survival fits use Efron Cox partial likelihood. Apparent fit statistics and loading ranks are exploratory and do not establish external predictive validity.

Examples

set.seed(2026)
X <- matrix(rnorm(80 * 6), 80)
colnames(X) <- paste0("G", 1:6)
time <- rexp(80, rate = exp(0.5 * X[, 1]))
delta <- rep(c(0L, 1L), 40)
fit <- fit_risdr_dual(
  X, time, delta = delta, response_type = "survival",
  sdr_method = "sir", cov_method = "oas", nslices = 4,
  dual_rank = 4, verbose = FALSE, d_max = 2, lambda_grid = c(0, 0.05)
)
fit$d
#> [1] 1
predict(fit, X[1:2, , drop = FALSE], type = "risk")
#>         1         2 
#> 0.9553409 0.4922246