Data Mining at IFS, TU Vienna

The Self-Organising Map is a popular unsupervised neural network model which has successfully been used for clustering various kinds of data.
The SOMVIS Package is an add-on for the Matlab SOMToolbox. It provides a graphical interface to access a set of visualisations, SOM quality measures, as well as clustering techniques such as k-means and Ward's linkage that can be applied on the SOM lattice.

Contact: Andreas Rauber.

Features

Visualisation

Besides the U-Matrix and Component Plane visualisations, which are already included in the Matlab SOMToolbox, the SOMVIS package additionally provides the following visualisations:

Metro Map
Gradient Field & Borderline
Neighbourhood Graphs (Graphical Methods)
P-Matrix
U*-Matrix
D-Matrix (variation of U-Matrix)

Quality Measures

The SOMVIS package additionally provides the following visualisations:

Intrinsic Distance
S. Kaski and K. Lagus. Comparing Self-Organizing Maps. In Proceedings of the International Conference on Artificial Neural Networks (ICANN '96), Bochum, Germany, July 16-19, pages 809-814, Berlin, 1996. Springer.
Topographic Error
K. Kiviluoto.. Topology preservation in Self-Organizing Maps. In Proceedings of the IEEE International Conference on Artificial Neural Networks (ICANN'96), pages 294-299. Piscataway, New Jersey, USA, June 1996.
Topographic Product
H. U. Bauer and K. R. Pawelzik. Quantifying the neighborhood preservation of Self-Organizing Feature Maps. In IEEE Transactions on Neural Networks, 3(4):570-579, July 1992.
Trustworthiness, Neighborhood Preservation
J. Venna and S. Kaski. Neighborhood preservation in nonlinear projection methods. An experimental study. In Proceedings of the Internationla Conference on Artificial Neural Networks (ICANN '01)pages 485-491. Berlin, 2001. Springer

Analytical Tools

We additionally provide tools to further analyse the data and maps, we provide a set of additional methods:

Map Clustering
To identify cluster boundaries, the SOM codebook (weight, model) vectors are clustered. SOMVIS provides the following methods:
- Linkage clustering, such as Single Linkage, Complete Linkage, and Ward's Linkage
- K-Means clustering
Data Projections
We provide visualisations of PCA, Sammons mapping and CCA projection methods.

License

The SOMVIS Matlab Visualisation Package for Self-Organising Maps is licensed under the GPL License, Version 3.0. This means you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version.

Installation & How-To

Download the package
Note: The SOMVIS Package builds on these other packages:
- Matlab SOM Toolbox (http://www.cis.hut.fi/projects/somtoolbox/)
- SDH Toolbox (http://www.oefai.at/~elias/sdh/download.html)
- The "dijkstra.m" file (http://www.mathworks.nl/matlabcentral/fileexchange/loadFile.do?objectId=5550&objectType=file)
All these packages are already included in the SOMVIS package.
Extract the package to a directory
Start Matlab, and navigate to the directory
Run "setPaths" to set the (relative) paths to the needed libraries
Note: When setting the Matlab path to include all the previously mentioned packages & files manually, the "somvis" directory has to be included above the "somtoolbox" directory, since a function has been overwritten.
Train a map
- Load a data set and train a SOM with it.
  Creating a SOM from own data can be done with the SOM Toolbox commands som_data_struct, som_normalize and som_make.
  See the Matlab SOMToolbox manual for more details.
- Some pre-trained SOMs (along with their data sets) are included in the SOM-VIS package, and can be loaded with the Matlab command: load datasetName
  The following datasets are included in the data/ directory:
  - Boston.mat
  - Cars93.mat
  - chickwts.mat
  - epil.mat
  - frac_big.mat
  - frac.mat
  - gilgais.mat
  - ionosphere_big.mat
  - ionosphere.mat
  - iris_big.mat
  - iris.mat
  - mtcars.mat
  - nlschools.mat
  - phonetic_big.mat
  - phonetic.mat
  - phonetic_reduced.mat
  - pluton.mat
  - quakes.mat
  - rock.mat
  - UScereal.mat
  - UScrime.mat
  - xclara.mat
  After loading, in the Matlab environment, two variables will appear: "data" and "map", of the internal SOM Toolbox format.
Start the GUI with somvis_gui (map, data)