Tucker Alternating Least Squares (HOOI) — tucker

Computes a Tucker decomposition of a dense tensor with target multilinear ranks via Higher-Order Orthogonal Iteration. Mirrors the MATLAB Tensor Toolbox tucker_als.

Usage

tucker_als(
  X,
  ranks,
  tol = 1e-04,
  maxiters = 50L,
  dimorder = NULL,
  init = "nvecs",
  printitn = 0L
)

Arguments

X: A Tensor or array-like object.
ranks: Integer vector of per-mode target ranks, or a scalar replicated across modes.
tol: Convergence tolerance on change in fit (default 1e-4).
maxiters: Maximum number of HOOI sweeps (default 50).
dimorder: Integer permutation giving the order in which factor matrices are updated.
init: Either "nvecs" (leading left singular vectors, default), "random" (random orthonormal), or a list of initial factor matrices.
printitn: Print fit every printitn iterations (0 to suppress).

Value

A TTensor.

Examples

set.seed(1)
X <- tensor(array(runif(60), dim = c(3, 4, 5)))
T <- tucker_als(X, ranks = c(2, 3, 3), maxiters = 20)