Improved ICA Decomposition for GOGARCH — improved_ica_decomposition • tsbs

Performs ICA decomposition with multiple restarts and quality diagnostics. Uses the RADICAL algorithm from tsmarch.

Usage

improved_ica_decomposition(
  residuals,
  method = "radical",
  n_components = NULL,
  n_restarts = 3,
  demean = FALSE,
  verbose = FALSE
)

Arguments

residuals: T x k matrix of residuals
method: Currently only "radical" is supported
n_components: Number of components to extract (default: k)
n_restarts: Number of random restarts (default: 3)
demean: Whether to demean the data (default: FALSE)
verbose: Print progress information

Value

List with components:

S: Independent components matrix (T x n_components)
A: Mixing matrix (k x n_components)
W: Unmixing matrix (n_components x k)
K: Pre-whitening matrix
method: "radical" or "pca_fallback"
quality: Quality metrics (see Details)
convergence_info: Restart diagnostics

Details

Improvements over direct radical calls:

Multiple restarts with best-fit selection based on negentropy
Quality metrics: independence score (target >0.8), reconstruction error (target <1\
Graceful PCA fallback if ICA fails

See also

estimate_garch_weighted_gogarch, gogarch_diagnostics