Factor Snob README file
-----------------------

Chris Wallace had been refining "Factor Snob" for some years
before his death in 2004, when he found time from his other
projects, including his book [1] (2005).
Some of us in his department had run it on several problems,
but Chris did not want to make it publicly available before
some more testing and tweaking. (He also wanted to brings its
file formats and command set into closer agreement with the
different program "Vanilla Snob", which he wrote in 2002.)
It is our belief that the code is essentially correct.

(See "examples/DataFormat" on Factor Snob's file formats.
 Also see http://www.allisons.org/ll/MML/Notes/SNOB-factor/
 for some notes on using Factor SNOB -- LA 2008.)

 +-------------------------------------------------------+
 | License                                               |
 | -------                                               |
 | 2008: Judy Wallace decided that "Factor Snob" be made |
 | available under the GNU General Public License, GPL   |
 | (http://www.gnu.org/copyleft/gpl.html).               |
 |                                                       |
 | Work using Factor Snob should cite Wallace [1],       |
 | Boulton & Wallace [3], and Wallace & Freeman [5].     |
 +-------------------------------------------------------+

About Factor Snob
-----------------

Snob does mixture modelling by Minimum Message Length (MML). Factor
Snob can handle correlated variables by finding single-factor models.
Finding such a factor can reduce the number of classes and
therefore shorten the message length of the model.

Factor Snob is also hierarchical. It explicitly constructs a tree,
taking the tree structure into the calculation for message length.


Short History of Snob
---------------------

Wallace & Boulton [2] gave a method for estimation of Multivariate
Mixture Models over Gaussian and Multinomial probability distributions.
It was inconsistent because it totally (rather than probabilistically)
assigned things (data) to components (clusters, classes).
Boulton and Wallace [3] considered hierarchical classification.
Wallace [4] made it consistent by introducing fractional assigment.
Wallace & Freeman [5] gave the MML theory of a single-factor model,
and Wallace & Dowe [6] the Poisson and von Mises distributions.

The MML book [1] is the best general reference on MML.


The MML Book
------------

[1] C. S. Wallace,
    Statistical and Inductive Inference by Minimum Message Length,
    Springer, isbn-13:978-0-387-23795-4, 2005.


Other References
----------------

[2] C. S. Wallace & D. M. Boulton,
    An Information Measure for Classification,
    Computer J., Vol.11, No.2, pp.185-194, 1968.

[3] D. M. Boulton and C. S. Wallace,
    An information measure for hierarchic classification,
    Computer J., Vol.16, No.3, pp.254-261, 1973.

[4] C. S. Wallace,
    An improved program for classification.
    Proc. ACSC-9, pp.357-366, 1986.

[5] C. S. Wallace and P. R. Freeman,
    Single Factor Estimation by MML,
    J. Royal Stat. Soc. B, Vol.54, No.1, pp.195-209, 1992.

[6] C. S. Wallace & D.Dowe,
    MML Mixture Modelling of multi-state, Poisson, von Mises circular and
    Gaussian distributions, 28th Symp. on the Interface, pp.608-613, 1997.

-------------------------------------------------------------------------------