SPARAFAC (Sparse PARAFAC) performs variable selection in high-dimensional multi-way models by inducing sparsity through L1 norm penalized minimization (Lasso). Uses Alternating Shrunken Least Squares (ASLS) with soft thresholding, enabling multidimensional co-clustering. Supports non-negativity and orthogonality constraints. Multiple random starts are recommended as the non-convex optimization may converge to local minima.
Morten Arendt Rasmussen [1]Rasmus Bro [1]
[1]Department of Food Science, University of Copenhagen, Denmark
