Thank you for your prompt answer. The breathing signal observations are the amplitude values as a function of time and phase. According to our model the hidden states are the different breathing types. Subjects, whose respiratiion process is regular, are likely to breathe, keeping the same cycle pattern/type, for many consecutive cycles. therefore dwelling in the same hidden state. The more regular the breathing process is, the more predictive its signal becomes the higher its amplitude autocorrelation order.
I guess my question is: can msm implement an AutoRegressive HMM ? It seems that depmixS4 can but it has a time series length constraint that I don't quite understand. Thank you in advance for your attention. Kind regards, Maura Edelweiss -----Messaggio originale----- Da: Walter Zucchini [mailto:[EMAIL PROTECTED] Inviato: mar 11/11/2008 11.32 A: [EMAIL PROTECTED] Oggetto: Re: R: Hidden Markov Models Dear Ms Monville, Hidden Markov models (HMMs), and that includes the msm implementation, are not based on the assumption that the observations are independent. Indeed HMMs are specifically designed to model serially dependent observations. Of course that doesn't mean that they can accommodate every type of serial dependence. It might turn out that HMMs are not useful for modelling whatever aspect of breathing you are investigating. HMMs are based on the assumption that the observations are "conditionally independent, given the states". This is a somewhat technical assumption that I won't try to explain by email, except to say that "conditional independence" does not imply independence of the observations themselves. Regards, Walter Zucchini -- Prof. Walter Zucchini, Institut fuer Statistik und Oekonometrie, Georg-August-Universitaet, Platz der Goettinger Sieben 5, 37073 Goettingen, Germany ----------------------------------------- Tel +49-551-397286 FAX +49-551-397279 ========================================= [EMAIL PROTECTED] wrote: > Dear Prof. Zucchini, > > I am reading the comprehensive on-line documentation about msm. > The positive side is that it seems it has been designed for biomedical > statistics, > like Clinical Trials. > The bad side is that it does not seem to model observations sequences that > are not > independent but instead are autocorrelated, as it is my case. I did not find > any mention to > correlated observations therefore I assume the authors did not have to face > this problem. > Did I get it wrong ? > > Since the breathing signals amplitude is an autocorrelated function of time > and phase, I would > greatly appreciate your comments about the possibility to use msm eventually > after carring out > some modifications if the source code is available. > > Thank you in advance for your attention. > > Kind regards, > Maura Edelweiss > > > -----Messaggio originale----- > Da: Walter Zucchini [mailto:[EMAIL PROTECTED] > Inviato: lun 20/10/2008 12.50 > A: [EMAIL PROTECTED] > Oggetto: Re: Hidden Markov Models > > Dear Ms Monville, > >> something in R that implements continuous HMMs > > The R-library "msm", "Multi-state Markov and hidden Markov models in > continuous time", might do what you want. > > Regards, > > Walter Zucchini > > > -- > Prof. Walter Zucchini, > Institut fuer Statistik und Oekonometrie, > Georg-August-Universitaet, > Platz der Goettinger Sieben 5, > 37073 Goettingen, > Germany > ----------------------------------------- > Tel +49-551-397286 FAX +49-551-397279 > ========================================= > > > [EMAIL PROTECTED] wrote: >> Dear Prof. Zucchini, >> >> My name is Maura Edelweiss. >> I am a physicist (just graduated from Washington University) with a genuine >> interest in Statistical Signal Processing. >> Dr. Lamb and I are trying to build a model of human breathing from some >> breathing signals. >> SSA and some extra analysis (R and C++ code ) show that there are only a few >> breathing cycle types. >> That is, humans breathe switching from one cycle type to another. >> The breathing process seems to be well modeled by a Continuous output >> Density Hidden Markov Model. >> Since neither of us has previous experience with HMMs, we wonder if there is >> something in R that implements continuous HMMs and is reasonably well >> documented. That might make it easier to get started. >> >> Thank you in advance for your attention and help. >> Kind regards, >> > > > > Alice Messenger ;-) chatti anche con gli amici di Windows Live Messenger e > tutti i telefonini TIM! > Vai su http://maileservizi.alice.it/alice_messenger/index.html?pmk=footer > Alice Messenger ;-) chatti anche con gli amici di Windows Live Messenger e tutti i telefonini TIM! Vai su http://maileservizi.alice.it/alice_messenger/index.html?pmk=footer [[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.
