search

mattjj / pyhsmm

MIT License
546 stars 173 forks source link

Use regular HMM and HSMM #63

Closed jmgo closed 8 years ago

jmgo commented 8 years ago

Hi,

this not really an issue, I just have a doubt about the capabilities of the package.

Does the package contains the optimization procedures for the regular HMM and HSMM? I want to know this, in order to be able to compare the results of the regular H(S)MM and HDP-H(S)MM.

Regards, jmgo

mattjj commented 8 years ago

It depends on what you mean by 'optimization procedures', but EM for both of those models is supported. The class names in models.py should make clear which classes correspond to which models.

jmgo commented 8 years ago

Ok, thanks for the answer, that was what I wanted to know.

0x10cxR1 commented 7 years ago

Hi @mattjj

can you provide more detailed tutorials of using the regular HSMM? Suppose there's no probability assumption.

Besides, in the code, I don't quite understand the parameters in the init of HSMMBase. I thought usually we should pass the initial state trans matrix, initial state-observation matrix and initial state probability, and initial duration probability matrix to it. I don't understand why there's an alpha, and what's init_state_concentration?

I appreciate it if you can explain more about it.

mathDR commented 7 years ago

@Zhijing1128 There is a paper on it: http://www.mit.edu/~mattjj/papers/uai2010.pdf

mattjj commented 7 years ago

My thesis has some more info and is a bit more cleaned up, particularly Ch 2 and Ch 3.

Unfortunately I can't write more of a tutorial right now, but take a look at the examples directory if you haven't already.