CP Decompositions
The Tensor Toolbox provides the following CP-like decompositions:
- cp_als - Alternating least squares (ALS) method, the gold standard
- cp_arls - Alternating randomized least squares (ARLS) method, efficient for massive dense tensors
- cp_als_lev - ARLS with leverage scores (ARLS-Lev), efficient for massive sparse and dense tensors
- cp_opt - Direct optimization (OPT) method
- cp_wopt - Weighted direct optimization (WOPT) method for handling missing data
- cp_apr - Alternating Poisson regression (APR) using KL-divergence fitting function for Poisson tensor decomposition
- gcp_opt - Generalized CP with alternative loss functions
- cp_sym - Direct optimization for symmetric (SYM) decomposition
- cp_isym - Direct optimization for implicit (I) symmetric (SYM) decomposition
- cp_spm - Subspace Power Method (SPM) for symmetric tensor decomposition
- cp_orth_als - Orthogonalized ALS (Orth-ALS), a simple modification of the ALS algorithm to periodically "orthogonalize" the estimates of the factors
- cp_nmu - Nonnegative CP using multiplicative updates (NMU)