Help SOM Toolbox -- Initialization & Training GUI

Purpose

A graphical user interface for initialization and training of Self-Organizing Maps.

Syntax

  som_gui([sD])

Optional input arguments

  sD          Training data.
     (struct) data struct
     (matrix) data matrix, size [dlen dim]

Help Topics

Introduction: What is SOM Toolbox -- Initialization & Training GUI ?
Load data: Load data.
From workspace
From file
Preprocess data: Preprocess
Initialize map: Change default values
Initialize
Train map: Change default values
Train
Save trained map: Save map
Visualize map: Visualize map
Restart from the beginning: Restart
Close: Close
Main window: Main window
Help: GUI help.
More information about SOM and SOM Toolbox.

Introduction

SOM Toolbox -- Initialization & Training GUI is Graphigal User Interface for initializating and training SOM. It is based on matlab functions in Somtoolbox vs.2.

Loading the data set

You can load the data set from main window by pushing the button LOAD DATA in upper left corner of window or by selecting Load Data from menu Load/Save. When the button is pushed or menu is selected, then the Load GUI is activated, see figure below. Data can be loaded from workspace or from the file. Select the radiobutton Workspace if you want to load data from workspace or select radiobutton File if you want to load data from file. If you selected File then select the appropriate file type from popupmenu, you can choose either dat file (SOMPAK format) or mat file, which is ordinary matlab .mat file. Write the name of data in upper editfield if you have chosen workspace or write the file name in another editfield below that. You can also push button Browse and then you see all the choises that you have in browse window or in workspace window. After that press button Load and selected data will be loaded and Load GUI disapear and control return back to Main GUI. Note that there is also one checkbox called Missing value char and one empty field related to that. If you select File and after that file type dat file you can redefine char, which is ignored in case of dat file, default value is tab

Figure 1. Load GUI

Preprocessing the data

After you have loaded the data then you can preprocess it. Select the menu called Tools from the menubar and from there select Preprocess Data and preprocessing GUI will be activated. After you have preprocessed data save it to the workspace and close the Preprocessin GUI after that you see little question dialog box, which askes the name saved data, default value is the name of data before preprocessing. After that the data is preprocessed and new default values are shown in main window.

Map initialization

In main window the static fields show default initialization parameters for the training data. if you press button INITIALIZE and map will be initialized according to these parameters. You can also change these parameters by pressing button Change values. After that Initialization GUI, see figure below, will be activated and you can change initialization parameters from there. Possible options are:

Type:: Type of initilailzation, either random or linear
Lattice:: Form of map, either rectancular or hexagonal
Shape:: Shape of map, sheet, cylinder or toroid
Map size:: Size of map in form vertical*horizontal map units
Munits:: Total number of map units, so that the proportio of horizontal and vertical map units is same as in default map size.

Press OK when you are changesd the parameters and the new parameters are set and the initialization gui will disapper and control returns to Main GUI. Press button INITIALIZE and map will be initilized according to new parameters.

Figure 2. Initialization GUI

Map training

After initialization you can train the map. Training have two phases, rough training and finetune phase. You can see default parameter for training in static fields of main window. General values are common both for the rough and finetune phase and varying values show spesific values for both two phases. You can train the map with these values by pushing TRAIN button. It is also possible do only the finetune phase, by pushing Only Finetune radiobutton down. You can also change these default values by pressing Change values button in botom of training part. After that you can choose training algorithm, either sequential or batch training, from the question box. After choosing the appropriate algorithm you can select parameters from parameter window, see below. The values you see first are the default values. Possible options are:

Mask:: BMU search mask, size dim*1
Neigh:: Neighborhood function, gaussian, cutgauss, ep or bubble,
Tracking:: Tracking level value between 0-3
Length type:: Training length type, either samples or epochs
Order:: Sample order, either ordered or random
Learning func:: Learning rate function, inv, linear or power
Radius initial:: Initial training radius
Radius final:: Final training radius
Alpha initial:: Initial learning rate
Training length:: Training length

NOTE, however that all of these values are not used in batch training and so you can not see or change these values. For example length type is allways epochs in batch training so this value is hidden from you and you cannot change it. You approve you choises by pressing button Set parameters. After that parameter window will disapper and control returns back to main window. Then you can press TRAIN button map will be trained.

Figure 3. Change training parameters GUI

Save trained map

After training you can save the trained map. Choose manu Load/Save from the menubar and there Save map and either Save in workspace or Write cod-file. If you choose Save in workspace, then question box asks the map name in workspace and the map is saved in that name. If you choose Save map, then browse window asks the name of file where the map is saved. The Map is written to .cod file, which is SOMPAK format.

Visualize trained map

You can visualize trained map by selecting Tools from menubar and then selecting Visualize map. After that Visualization GUI is activated and you can visualize map.

Start from beginning

If you want to start from the beginning, this might be handy if you make some mistake or you want to train another map, then the best way to do it is by selectin Tools from the menu bar and Clear all from there.

Close the program

You can close the program when ever you wish by pressing CLOSE button in main window or by selecting Tools from menubar and Close Figure from there.

Main window

The main window has three separate parts, which are Information, Initialization and Training. Information part include useful information about map, data and actions which have been carried out. The Status field contain information about the last action which have been carried out. The map field contains the name of map and the data field contain the name of training data.

Initialization part contain all the information and actions needed to initialization of map. The static fields show the default initialization values for the particular data set. It is possible to change these values by pushing the button Change values in bottom of initialization part.

Training part contains all the information and actions needed to train the map. Also in this case, static fields show default training values. Genenral values are common for both rough and finetune training. Varying values are separate for these two phases. It is possible change these values by pushing the button Change values in bottom of training part. There is also radiobutton only finetune, when this button is activated the Rough values change inactive and only finetune is carried out.

Figure 4. Main GUI

Help

There is menu called Help/Info in main window where you can find help. WWW Help shows this page and Help window shows this page in ascii format.About GUI shows some general information about this GUI in help window. If you didn't find answer to your problem then you can try to look your answer in following web pages.
Back to begining