Hidden Markov Model by Matrix and Tensor Decomposition // Solves Hidden Markov Models (HMMs) via matrix and tensor decomposition. Converts observation sequences to co-occurrence matrices/tensors and applies Symmetric Non-negative Matrix Factorization (symNMF), Singular Value Decomposition (SVD), CANDECOMP/PARAFAC (CP) decomposition, or Tensor-Train (TT) decomposition to recover HMM parameters. Also provides standard HMM algorithms (Forward, Backward, Viterbi, Baum-Welch) for comparison. The spectral learning approach for HMMs is based on Hsu, Kakade, and Zhang (2012) doi:10.1016/j.jcss.2011.12.025. The symNMF method is described in Kuang, Yun, and Park (2015) doi:10.1007/s10898-014-0247-2. The Tensor-Train decomposition is described in Oseledets (2011) doi:10.1137/090752286.