Note: Elias Pampalk is not working at OFAI anymore. These pages are no longer updated and kept for archiving purposes only. For any further information, please visit Elias Pampalk's current personal webpage.

 Last Updated: 05.07.2002


The GHSOM Toolbox contains several functions to train, visualize and label a GHSOM. Most functions rely on some functions of the SOM Toolbox. For details on each function try typing "help <function>" or for full details try "type <function>" after installing the toolbox, in the Matlab environment.

The functions of the GHSOM Toolbox are categorized in:


The main training function is "ghsom_train". This function calls all other necessary training functions as needed. The other training functions are "ghsom_train_grow", "ghsom_train_grow_orientation", and "ghsom_train_grow_sampleinit".
There are several options available such as the orientation of sublayers, the initialization of the different layers, and the type of error calculated.


There are basically three options for labeling a GHSOM. The first "ghsom_datalabels" labels each map unit with the names of the data items. The "ghsom_labelsom" and "ghsom_websom" functions label the map units based on the attributes they represent.


There are two visualization functions which normally would be used subsequently. The first function "ghsom_visualize_grid" draws the basic GHSOM structure and adds coloring (e.g. component planes or hierarchy level) if desired.
The second function "ghsom_visualize_labels" adds the map labels on top of the ghsom grid. Alternatively, the model vectors can be visualized - if they represent 2-dimensional time series data - using "ghsom_visualize_2dts".

External Interface

All Matlab data structures can be stored using the Matlab commands "save <filename> <variable1> <variable2> ..." (and "load <filename>" to load stored variables). An external interface for the SOM_PAK data format is supplied by the functions "ghsom_write_cods" and "ghsom_read_cods".


To demonstrate the usage and some of the possibilities a few demonstrations are supplied with the toolbox. The functions are "ghsom_demo", "ghsom_demo1", "ghsom_demo2", "ghsom_demo3", and "ghsom_demo4". For details on how the demonstrations call other functions use "type <function>". The demonstrations with documentation are available on this web page.


There are 2 functions which are used by other functions, namely "ghsom_idx2data" and "ghsom_map_data". The first is needed to obtain the sub dataset belonging to a sub-map of GHSOM structure. The second is necessary when reading data from the SOM_PAK interface.
Furthermore, a test dataset is supplied which has been published by Kohonen and contains some animals ("animals.dat"), a dataset consisting of 77 pieces of music ("music.mat"), and a dataset representing musical performances of Chopin pieces ("worms.mat").