HMM : Learning states in a signal. (using hmmlearn python library)

Asked Dec 08 '19 at 15:57

Active Dec 13 '19 at 11:40

Viewed 232 times

I have time series data for which I am trying to learn 3 states in my HMM model. The output which I am getting is this. I want to get three separate states for the 3 distinct clusters of data points. But the first cluster has mixed states learned from HMM instead of a single state.

Here is my code for the same.

 model = hmm.GaussianHMM(n_components=3, covariance_type="full", verbose = True, n_iter =1000)
 model.fit(train_array1.reshape((-1,1)))
 Z1 = model.predict(test_array1_s.reshape(-1,1))  
 p = model.predict_proba(test_array1_s.reshape(-1,1))  
 plt.plot(p)

The plot of probabilites of each state for the test data is this.

How can I get three distinct states for the the data?

edited Dec 08 '19 at 20:49

asked Dec 08 '19 at 15:57

viswarup misra

What is the question? – Mert Gülsoy Dec 08 '19 at 18:04
How can I get three distinct states for the the data? – viswarup misra Dec 08 '19 at 20:49
What are your input features? Have you checked that they are predictive of the state? Tried any feature engineering? – Jon Nordby Sep 11 '20 at 06:25
Is there an alternating pattern in your data? It is a bit hard to see from the plot. Can you plot it wider, so that the sequences of dataspoints are more clear? – Jon Nordby Aug 05 '22 at 19:48

HMM : Learning states in a signal. (using hmmlearn python library)

0 Answers0