In Signal processing, there is a topic of 'Quantization' (the process of mapping input values from a large set to output values in a (countable) smaller set ('states') ). I would like to construct a Markov Chain by relating the states these different 'states' interact with each other and the probability of these states coming about.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
5
Regime detection with hidden Markov model: http://scikit-learn.sourceforge.net/stable/modules/hmm.html
-
1$\begingroup$ Thank you so much. This was pretty much what I was looking for, however it says that this has been depreciated in Scikit. Can you guide me if Scikit has some current implementation tutorials for regime detection of Time Series analysis? $\endgroup$ Commented Jan 29, 2018 at 19:22
-
$\begingroup$ Statsmodels also provide a library of Markov-switching models: statsmodels.org/dev/examples/notebooks/generated/… statsmodels.org/dev/examples/notebooks/generated/… Seems that you may need to modify the examples a bit for it to work though. But worst case scenario, you should be able to still get their code and adjust accordingly. Out of my curiosity, what's the usage in your case? $\endgroup$– LiptonCommented Jan 29, 2018 at 20:22
-
$\begingroup$ I want to use them to "discretise" a time series in different "states" so that I can use it in a Dynamic Bayesian Network (in pgmpy). $\endgroup$ Commented Jan 29, 2018 at 21:05
-
$\begingroup$ I'm not understanding how to use the links for regime detection. That is the issue I am facing right now. $\endgroup$ Commented Jan 29, 2018 at 21:27
-
$\begingroup$ By any chance, are you saying to use the probability of every state and then use their intersection (or maximum) to determine which state the graph is in? $\endgroup$ Commented Jan 29, 2018 at 23:08