R help - Nov 2008 - choice of an HMM package

We are trying to build a human respiration model.
Preliminary analysis of some breathing signals has shown that humans breathe
through switching among
a finite number of patterns.
Hidden Markov seems to be the right approach. Since most of our code is
written in R scripting language, finding an R package implementing an HMM
that we can use for our prototype would be very helpful.
I have been suggested both *msm* and *depmixS4.*
I have no previous experience with HMMs and feel at a loss about making a
sensible choice.
As a novice I am more attracted by msm because of the comprehensive
documentation, with worked out examples, that I am printing out. Whereas I
could not find anything but the usual R function call description for
depmixS4.
Moreover, I cannot make a sense of depmixS4 documentation mentioning time
series of length 1 ... Does it mean depmixS4 models time series made up of
one single observation ? Sorry for my trivial question.

In my case I have to model a variable which is an autocorrelated continuous
function of time. The transition probabilities may as well depend on time.
Therefore I believe we need an AutoRegressive Continuous Density HMM,
possibly also non-stationary.
Any suggestion from HMM experts is welcome.

Best regards,

-- 
Maura E.M

	[[alternative HTML version deleted]]

Dear Maura,

On 9 Nov 2008, at 02:24, Maura E Monville wrote:
> We are trying to build a human respiration model.
> Preliminary analysis of some breathing signals has shown that  
> humans breathe
> through switching among
> a finite number of patterns.
> Hidden Markov seems to be the right approach. Since most of our  
> code is
> written in R scripting language, finding an R package implementing  
> an HMM
> that we can use for our prototype would be very helpful.
> I have been suggested both *msm* and *depmixS4.*
> I have no previous experience with HMMs and feel at a loss about  
> making a
> sensible choice.
> As a novice I am more attracted by msm because of the comprehensive
> documentation, with worked out examples, that I am printing out.  
> Whereas I
> could not find anything but the usual R function call description for
> depmixS4.
?depmixS4 provides a brief overvied of the package
?depmix  tells you how to specify models
?fit tells you how to fit specified models

The package also includes a vignette depmix-intro.pdf with some worked
examples.
> Moreover, I cannot make a sense of depmixS4 documentation  
> mentioning time
> series of length 1 ... Does it mean depmixS4 models time series  
> made up of
> one single observation ? Sorry for my trivial question.
As the documentation says depmixS4 also fits non-dependent mixtures,  
however that
does not seem applicable in your case as you want to model time  
series with HMMs.
There is no principled limitation in the lengths of timeseries  
depmixS4 can handle.
> In my case I have to model a variable which is an autocorrelated  
> continuous
> function of time. The transition probabilities may as well depend  
> on time.
> Therefore I believe we need an AutoRegressive Continuous Density HMM,
> possibly also non-stationary.
Can you provide a reference for this kind of model as that will make  
your question
much more clear.
> Any suggestion from HMM experts is welcome.
>
> Best regards,
>
> -- 
> Maura E.M
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting- 
> guide.html
> and provide commented, minimal, self-contained, reproducible code.
Ingmar Visser
Department of Psychology, University of Amsterdam
Roetersstraat 15
1018 WB Amsterdam
The Netherlands
t: +31-20-5256723



	[[alternative HTML version deleted]]