DELTA LATENT DIRICHLET ALLOCATION (DeltaLDA)
Version 0.1.1

David Andrzejewski ([email protected])
Department of Computer Sciences
University of Wisconsin-Madison, USA

This software implements the DeltaLDA model [1] for discrete count
data.  DeltaLDA is a modification of the Latent Dirichlet Allocation
(LDA) model [2] which uses two different topic mixing weight priors to
jointly model two corpora with a shared set of topics.  The inference
method is Collapsed Gibbs sampling [3].  This code can also be used to 
do "standard" LDA, similar to [3] (see example code).

This code implements DeltaLDA as a Python C extension module.


BUILD/INSTALL

Building this module requires installation of Python and NumPy, as
well as a C compiler.  From the command-line, do:

% python setup.py install

(Note that if NumPy or Python are installed to non-standard locations,
you may need to make the appropriate changes in setup.py)

You can then test the installation with:

% python test/TestDeltaLDA.py

There is also a simple example scipt showing how to use deltaLDA:

% python example.py


LOCAL INSTALL

If you do not have write access to your Python installation directories,
you will need to tell setup.py to install this module somewhere else.
For example:

% python setup.py install --prefix=~/local

will install the module under a subdirectory of your home directory called 
"local".

It may then be necessary to let Python know where that is by setting
the PYTHONPATH environment variable (e.g., in .bashrc or .cshrc).  For
our example this might involve adding something like the line:

setenv PYTHONPATH ~/local/lib/python2.5/site-packages


HOW TO USE

The commenting in the example.py script explains the meanings and
types of all input and return arguments.  For more detailed examples
of good/bad inputs, you can examine the tests in TestDeltaLDA.py.


LICENSE

This software is open-source, released under the terms of the GNU
General Public License version 3, or any later version of the GPL (see
LICENSE.txt).


REFERENCES

[1] Andrzejewski, D., Mulhern, A., Liblit, B., and Zhu, X. (2007). 
Statistical debugging using latent topic models. 
In Proceedings of the 18th European Conference on Machine Learning
(ECML), 6-17.

[2] Blei, D. M., Ng, A. Y., and Jordan, M. I. (2003). Latent dirichlet
allocation.  Journal of Machine Learning Research (JMLR) 3
(Mar. 2003), 993-1022.

[3] Griffiths, T., and Steyvers, M. (2004).  Finding Scientific
Topics. Proceedings of the National Academy of Sciences (PNAS), 101
(suppl. 1), 5228-5235.


VERSION HISTORY
0.1     Initial release
0.1.1   Fixed some npy_intp[] memory leaks
        Fixed bug where betasum was taken over *wrong dimension* of
        beta, causing phi to not be normalized correctly