Parallel Analysis via time-wise permutation

Computes a null distribution for the leading singular value by permuting time indices independently within each column and selecting k where observed singular values exceed the (1 - alpha) quantile of the null.

Usage

choose_rank_pa(W, B = 100L, alpha = 0.05)

Arguments

W

Whitened matrix (T x V).

B

Integer number of permutations (default 100).

alpha

Significance level (default 0.05).

Value

A list with k, threshold (null quantile), null_s1, and s.

See also

choose_rank_gd() for Gavish-Donoho approach, blocked_cv_recon_error() for cross-validation approach

Other rank selection: blocked_cv_recon_error(), choose_rank_gd()

Examples

# \donttest{
set.seed(1)
W <- matrix(rnorm(200), nrow = 20, ncol = 10)
pa <- choose_rank_pa(W, B = 20, alpha = 0.05)
# }