Added spectral learning for categorical HMMs. Fixes #283

[mikecondon]

Added spectral learning for categorical HMMs as described in #283. This implementation followed

Anandkumar, Animashree, et al. "Tensor decompositions for learning latent variable models." Journal of machine learning research 15 (2014): 2773-2832.

In utils/utils.py implemented:

• the robust tensor power method.
• low rank pseudoinverse. Used truncated SVD to find the pseudoinverse.
• multilinear map.

In hidden_markov_model/models/categorical_hmm.py, implemented:

• sample cross moment to find a given symmetric orthogonally decomposable view of the n order cross moments.
• fit_moments to find the parameters of the generating hmm.

In ssm.py, added:

• minimal code to allow for future expansion of spectral learning to other HMM variants.

[mikecondon]

Hi there! I’m checking back in on this PR. It looks like the automated workflows are currently awaiting approval from a maintainer to run. Once those are cleared, I’m happy to address any feedback or CI failures that might pop up. Thanks!