Error with HMM analysis

[ffavelar]

I'm trying to run the tutorial of the Hidden Markov Model in Ubuntu 20.04.6 LTS (GNU/Linux 5.15.0-75-generic x86_64) and Python 3.9.10. I get this output:

100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 51/51 [00:01<00:00, 29.28it/s]
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 51/51 [00:01<00:00, 46.18it/s]
0%| | 0/3 [00:00<?, ?it/s]
MultinomialHMM has undergone major changes. The previous version was implementing a CategoricalHMM (a special case of MultinomialHMM). This new implementation follows the standard definition for a Multinomial distribution (e.g. as in https://en.wikipedia.org/wiki/Multinomial_distribution). See these issues for details:
hmmlearn/hmmlearn#335
hmmlearn/hmmlearn#340
0%| | 0/3 [00:00<?, ?it/s]
Traceback (most recent call last):
File "HiddenMarkovModel.py", line 370, in
model = model.fit(
File "/home/jorge/.local/lib/python3.9/site-packages/hmmlearn/base.py", line 469, in fit
self._check()
File "/home/jorge/.local/lib/python3.9/site-packages/hmmlearn/hmm.py", line 907, in check
raise ValueError(
ValueError: emissionprob must have shape (n_components, n_features)

I understand that emissionprob_ does not have the proper shape, but I don't know how to fix the code. Any help will be appreciated.

[p-j-smith]

Hi @ffavelar, thanks for raising this issue! It looks like MultinomialHMM has been renamed to CategoricalHMM, so in your script HiddenMarkovModel.py you will need to find the line:

model = hmmlearn.hmm.MultinomialHMM(
        n_components=n_hidden_states,
        init_params='',
        verbose=True,
        n_iter=100  # maximum number of iterations
    )

and change it to:

model = hmmlearn.hmm.CategoricalHMM(
        n_components=n_hidden_states,
        init_params='',
        verbose=True,
        n_iter=100  # maximum number of iterations
    )

Then hopefully everything should still work as before, but let me know if you have more issues

[ffavelar]

Hi, thank you for the reply. But now I get to the following error:

Traceback (most recent call last):
File "HiddenMarkovModel.py", line 290, in
emission_probs = get_emission_probs(
File "HiddenMarkovModel.py", line 252, in get_emission_probs
state_counts[state] = np.sum(states == state, axis=0)
ValueError: could not broadcast input array from shape (51,) into shape (204000,)

[p-j-smith]

so the issue is with this line in your script:

state_counts[state] = np.sum(states == state, axis=0)

inside the function get_emission_probs. However, looking at the tutorial notebook I can't see this line in this function - is it something you've added?

[ffavelar]

Sorry, I forgot that I messed a little with the code before contacting you. Now everything is working great. Thank you.

[p-j-smith]

No worries, glad it's working for you now 🙂