[6572] in s-news-athena
Enhanced version of the Gibbsit program
daemon@ATHENA.MIT.EDU (slewis@stat.washington.edu)
Mon Mar 6 18:57:48 1995
Date: Mon, 6 Mar 95 15:36:14 PST
From: slewis@stat.washington.edu
To: bayes-news@stat.cmu.edu, s-news@utstat.toronto.edu
GIBBSIT: New Version
--------------------
Gibbsit is a program to determine how many iterations are needed in a
Gibbs sampler or Markov chain Monte Carlo (MCMC) algorithm, based on a
pilot sample. It also estimates the number of iterations needed for
convergence and diagnoses convergence.
New features
------------
The new version is much enhanced and has the following additional features:
* Results can be given for several parameters or variables and several
quantiles at once.
* The number of iterations to be skipped for approximate independence
is returned.
* Results can be given for estimating a probability (the previous version
gave results only for estimating quantiles of a variable).
* The convergence diagnostic of Raftery and Lewis (1992b) (denoted by I)
is computed.
* Several bugs have been fixed.
* An S (or S-PLUS) version is now available, in addition to the Fortran
version, written jointly with Karen Vines.
* An XLISP-STAT version is available, written by Jan de Leeuw.
* Background papers that give details of the methods are available by
ftp or over the World Wide Web (details in the software).
Availability
------------
To get the new version of the Fortran program from Statlib send the message
"send gibbsit from general" to statlib@stat.cmu.edu
To get the S implementation of gibbsit from Statlib send the message
"send gibbsit from S" to statlib@stat.cmu.edu
To get Jan de Leeuw's XLISP-STAT version of gibbsit, you can access URL:
ftp://ftp.stat.ucla.edu/pub/lisp/xlisp/xlisp-stat/code/statistics/simulation/mcmc
on the World Wide Web,
or by regular ftp using the following commands:
ftp ftp.stat.ucla.edu
login as anonymous
enter your email address as your password
ftp> cd pub/lisp/xlisp/xlisp-stat/code/statistics/simulation/mcmc
ftp> get gibbs.lsp
ftp> quit
References
----------
Raftery, A.E. and Lewis, S.M. (1992a). How many iterations in the Gibbs
sampler? In Bayesian Statistics 4 (J.M. Bernardo et al., eds.),
Oxford University Press, pp. 763-773.
Raftery, A.E. and Lewis, S.M. (1992b). One long run with diagnostics:
Implementation strategies for Markov chain Monte Carlo.
Statistical Science 7, 493-497.
Raftery, A.E. and Lewis, S.M. (1995). The number of iterations, convergence
diagnostics, and generic Metropolis algorithms. To appear in
Practical Markov Chain Monte Carlo (W.R. Gilks, D.J. Spiegelhalter and
S. Richardson, eds.), London: Chapman and Hall.
-----------------------------------------------------------------------
Raftery and Lewis (1995) is available as a PostScript file that you can
download from the World Wide Web or by ftp, and then print out.
You can get it from the Web using URL:
http://www.stat.washington.edu/tech.reports
It is also available via regular ftp using the following commands:
ftp ftp.stat.washington.edu (or 128.95.17.34)
login as anonymous
enter your email address as your password
ftp> cd pub/tech.reports
ftp> get raftery-lewis2.ps
ftp> quit
Steven Lewis (slewis@stat.washington.edu)
