Support for HMMs with num_states=1

I am trying to fit various HMM classes (LinearRegressionHMM, or GaussianHMM) to my data but it does not let me pass num_states=1. For num_states > 2, everything works as expected. I wanted to know whether no support for num_states=1 is the intended behavior.

It's easy enough to write code for simple linear regression outside dynamax, however, it still makes the comparison with num_states>2 cases error-prone (as one might be using different constants in log-likelihood calculations, etc.).

If it helps, the error occurs while trying to initialize the Dirichlet distribution.

File "/Users/us/project/fitting.py", line 20, in fitEM
    params, props = hmm.initialize(key)
  File "/Users/us/dynaenv/lib/python3.10/site-packages/dynamax/hidden_markov_model/models/gaussian_hmm.py", line 649, in initialize
    params["initial"], props["initial"] = self.initial_component.initialize(key1, method=method, initial_probs=initial_probs)
  File "/Users/us/dynaenv/lib/python3.10/site-packages/dynamax/hidden_markov_model/models/initial.py", line 45, in initialize
    initial_probs = tfd.Dirichlet(self.initial_probs_concentration).sample(seed=this_key)
...
ValueError: Argument `concentration` must have `event_size` at least 2.

Start in hidden_markov_model/models/initial.py at the Dirichlet initialization called from hidden_markov_model/models/gaussian_hmm.py:649. Trace how initialization handles num_states=1, then verify that LinearRegressionHMM and GaussianHMM can initialize and fit with one state without the reported concentration error.