🚀 go-pugleaf

I have some data that tracks behaviour at monthly intervals over a period
of 30 months.  At each month, the status of each case is measured on
50'ish variables.  In the past, this type of data (not this particular
data set) has been modelled as a Markov system.  The Markovian states were
defined in terms of conjunctions of subranges of 4 or 5 of the variables.
 The identification of the (near enough to) Markovian states was carried
out by an ad hoc process pushed along by a bit of a priori intuition about
the important variables.

** I would be immensely grateful if someone could point me at an algorithm
or two for systematic identification of the Markovian states defined in
terms of the observed variables. **

** Extra credit question: The process is almost certainly not stationary,
due to seasonal effects.  Does anyone have any pointers to work dealing
with Markov systems in which the transition probability matrix has a
constant pattern (whatever that may mean) over time, but the absolute
values in the matrix vary smoothly as a function of a season variable? **

Thanks in advance,
Ross

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet