------------------------------
COPYRIGHT AND LICENSE
------------------------------

The copyright and license information for this software are detailed in file license_Nikolaos_Gkalelis.txt.

This software re-uses certain pre-existing parts of code; additional licenses apply to these software parts.
For details on licence and copyrights concerning these software parts see files license_Michael_Chen.txt, PCA.m, sqdistance.m, logdet.m.


------------------------------
HOW TO USE THE SOFTWARE
------------------------------

The function "learn_msda.m" and "test_msda.m" are central for using the software.
The function "learn_msda.m" receives as input a set of parameters concerning the training set and learns the MSDA parameters for this dataset (e.g., MSDA projection matrix, etc.).
It iteratively calls the functions "selectClassToRePartitionPost.m", "incrementSubclassesOfClass.m" and "cmp_msda_mat.m" to identify the best MSDA parameters.
The actual nongaussianity of each subclass partition is computed using the function "negEntropyIncrement.m".
New subclass partitions are computed using an extended kmeans algorithm that utilizes a deterministic initialization and creates balanced subclasses.
Finally, the function "test_msda.m" receives as input the training set, the learned MSDA parameters and the test set, and exploits the nearest neighbor classifier for classifying the test observations.

A comprehensive description concerning the functionality of each matlab function is provided in the header of the respective m-file.

For more details on using this software please refer to "notebook_msda.m" script, where an example application of this software is provided utilizing the Wisconsin Breast Cancer Data Set of UCI repository.


------------------------------
RELATED PUBLICATIONS
------------------------------

If you find this software useful in your work, please cite the following publications:

1. N. Gkalelis, V. Mezaris, I. Kompatsiaris, "Mixture subclass 
discriminant analysis", IEEE Signal Processing Letters, vol. 18, no. 5,
pp. 319-322, May 2011

2. N. Gkalelis, V. Mezaris, I. Kompatsiaris, T. Stathaki, "Mixture
subclass discriminant analysis link to restricted Gaussian model and
other generalizations", IEEE Transactions on Neural Networks and Learning 
Systems, vol. 24, no. 1, pp. 8-21, January 2013.


------------------------------
COMMENTS/QUESTIONS
------------------------------

Please send any comments and questions to gkalelis@iti.gr
