MC2Pack  -- General comments
-----------------------------
MC2Pack is an ox package meant to help in sampling using a range of
Bayesian sampling methods. Implemented are 
  MC_MH     Metropolis Hastings
  MC_IS     Importance Sampling
  MC_APS    Adaptive Polar Sampling
  MC_APIS   Adaptive Polar Importance Sampling
  MC_GG     Griddy Gibbs Sampling

Sampling can be done specifying not much more than the posterior density
itself. For the Gibbs sampler, the conditional densities are by default
constructed numerically, though for an increase in speed analytical
densities can be used as well. 

The APS/APIS samplers (from the family of Adaptive Radial-based
Direction samplers, see Bauwens, Bos, Van Dijk and Van Oest, Journal of
Econometrics, 2004) are sampling methods which are more robust against
multimodality and heavy tails than standard MH or GG algorithms.

MC2Pack  -- Installation
-------------------------
To install MC2Pack, unzip the installation file from the ox main
directory. As it depends on the packages OxUtils and GnuDraw, install
those as well by extracting the files oxutils.zip and gnudraw.zip in the
same manner.

Documentation is included in HTML format, and in PDF. The html document
is most easily accessed by adding a line 
       <tr><td><a href="../packages/mc2pack/doc/mc2pack.html">MC2Pack reference</a></td></tr>
to <ox-home>/doc/oxmenu.html, eg after the line referring to the
graphics reference.

MC2Pack  -- Usage
------------------
In the packages/mc2pack/doc directory a pdf file is provided with
explanation on the sampling methods provided, and an example is
discussed. The file discussed is 
  packages/mc2pack/samples/mcstack_clean.ox
In 
  packages/mc2pack/samples/mcstack_clean_aug.ox
a clean implementation of the augmented Gibbs sampler is given, whereas 
  packages/mc2pack/samples/mcstack.ox
is a more elaborate version of the program allowing command line
parameters controlling the type of sampling, the size, the convergence
criteria etc. E.g.
  oxl mcstack m 4 s 1000 aug
would sample using the GG sampling (MH=0, GG= 4), sample size 1000, and
applying the data augmentation.   
  oxl mcstack m 4 nb s 1000 aug
does the same thing, but not using the analytical conditional density of
beta (nb). The full set of options is
  nogr      skip drawing graphs
  m 0       method, 0=mh, 1=is, 2=aps, 3=apis, 4=gg
  s 1000    size of final sample, default= 1000
  f 0.9     mahalanobis fraction, 0.9 is default
  rw 0.2    use rw candidate, with fraction of covariance
  nb        skip beta conditional density
  ns        skip simulation
  aug       use augmented Gibbs sampling
  opt       optimize posterior at start
  base mcs  output base
  seed 1234 set random seed

The package is still under development; some function calls might get
changed over time, and bugs hunted down. If you happen to find any,
please let me know at cbos@feweb.vu.nl

Charles Bos, 22/6/2004
http://www.tinbergen.nl/~cbos  
  
