NOTES ON THE VERSION OF 1998-09-01

Changes in this version:

1) A new gibbs-ext-indicators operation has been added for mixture
   models.  This is an alternative to the met-indicators operation
   for infinite models.  The "no gaps" method of MacEachern and
   Mueller is also implemented as a special fudge on this.

2) Added 't', 'o', and 'h' quantities for mixture models.  For details,
   see mix-quantities.doc.

3) The mean-prior argument for mix-spec is now optional.  When absent,
   it is effectively zero.  See mix-spec.doc for details.

4) Reference is now made to my tech report on "Markov chain sampling
   methods for Dirichlet proces mixture models", which describes the
   sampling methods for mixture models used in this software.  The
   'ex-mix' directory contains the command files for the tests done 
   in that paper.

5) Added an example of implementing a random effects model using the
   facilities of the 'dist' module (see Ex-bayes-e.doc).

6) The 'ex-ais' directory contains command and data files used for the
   tests in my paper on "Annealed Importance Sampling" (revised version
   of 31 August 1998).

7) Added a "W" quantity to what can be displayed by mc-ais.  This 
   is more useful than "V" when there are a few very bad runs.  See
   mc-ais.doc for details.

8) The maximum number of temperatures in a tempering schedule really
   has been increased to 2001 from 1001.  (This was claimed to be so
   for the last release, but wasn't actually.)  Old log files that 
   used the mc-temp-sched command won't be readable by this release.

9) Various improvements were made to the documentation.

10)Changed the way the tar file is built so that untarring puts all
   the files and directories inside one directory with a name like
   fbm.YYYY-MM-DD.


Bug fixes.

Fixed a bug in the met1-indicators operation when applied to mixture
models with a finite number of components.  Fixed a bug that prevented
ExpGamma and ExpGamma2 from being used with function rather than
distribution syntax.  Fixed some minor problems with error checks.


Known bugs and other deficiencies.

1) The facility for plotting quantities using "plot" operations in xxx-mc
   doesn't always work for the first run of xxx-mc (before any
   iterations exist in the log file).  A work-around is to do a run of
   xxx-mc to produce just one iteration before attempting a run of
   xxx-mc that does any "plot" operations.

2) The CPU time features (eg, the "k" quantity) will not work correctly
   if a single iteration takes more than about 71 minutes.

3) The latent value update operations for Gaussian processes may recompute 
   the inverse covariance matrix even when an up-to-date version was 
   computed for the previous Monte Carlo operation.

4) Covariance matrices are stored in full, even though they are symmetric,
   which sometimes costs a factor of two in memory usage.