README

This is the accompanying code for Raffa JD and Dubin JA (2014):
"Multivariate Longitudinal Data Analysis with Mixed Effects Hidden Markov Models"

For updates please check:  http://github.com/jraffa/MHMM


We've included two files (in addition to this one):

ff-final.R: includes library functions including some R/C++ code (version 0.002).
This includes a function to generate simulated (genSim) data from the models considered in the paper
in addition to a Gibbs sampling algorithm for the approach outlined in the paper (simulated).
Both functions are documented in this file, but it's probably easiest to start with the example file:

example.R: has an example using both functions.


These functions depend on a number of R libraries, all of which are available at CRAN:

Rcpp, RcppArmadillo, inline, MCMCpack, and mvtnorm.

If you run this on a Windows machine you should download R-tools, which include a C++ compiler that will compile the embedded c++ code.
The process should be entirely automated, and we have tested this on both Linux and Windows.  For reference, we were using:

Rcpp version 0.11.2
RcppArmadillo version 0.4.300.8.0
inline version 0.3.13 
MCMCpack version 1.3-3
mvtnorm version 0.9-99992

all on R version 3.1.0 or 3.1.1 with R tools 3.1 for Windows.


Keep in mind this code is still very experimental.  Use with caution!
It can also take quite a long time to run.
In our example using the smoking cessation example, for 354 smokers, each
measured at most at six time points, we typically ran the MCMC algorithm for >3M iterations.
This can easily take the better part of 24 hrs.

We will continue to develop this code at: http://github.com/jraffa/MHMM
so please check there for updates, patches, and extensions.

Please e-mail questions to: jraffa@uw.edu and make sure to indicate what version of this software you are using,
 and include other relevant R information (version, package versions, etc).