SOG/ISOM Applet Usage

What the heck is this applet doing?

If you want to see other canned examples you can load some by entering a valid example name in the "input" field and hitting the "load" button. Current examples are:

• circle
• double-ring
• double-square
• double-triangle
• example3
• example4
• free-tree
• K6-fullscreen (will only work if your browser/appletviewer allows to resize the applet window to aprox. 1000 x 700)

Beware: If you don't understand how to set the applet parameters properly, it is quite likely that it will not really do what you expect it to do...
So it might be a good idea to continue reading.

Full Instructions

Here is a snapshot of the applet

Entering a new graph can be done in two ways: you can either draw a graph manually or load it from a file. Since this is a Java 1.0.X applet you can only load a file that resides on the server, so you might have to download the applet to use this function.

While the algorithm is running, some colors are used to indicate the current role of the nodes. The small jumping black dot indicates the current input stimulus. The winner node is colored red and the topological neighborhood that is updated is colored brown. Yellow nodes remain unchanged in the current epoch.

The controls in detail:

• Graph-Editor Modes: The check-boxes on the right-hand side control the interaction mode of the graph editor. Their function should be obvious.
  • In +node mode a new graph node will be generated with every mousclick in the layout area. The node label will be the text in the input field at the time of the mouse-click.
  • In -node mode every node on which a mouse click is issued will be deleted.
  • In +edge mode new edges are drawn. The mouse must first clicked over the "from" node, then over the "to" node. (Currently all edges are undirected).
  • In -edge mode edges can be deleted by first clicking on the one adjacent node, then on the other node.
  • The area mode is different. It allows to define the active area for input stimuli according to the shape of the input area currently selected in the controls window. This is simply done by dragging the mouse while pressing button-1. Caution: Pressing the initialize button will reset the area to the largest area possible. When this mode is selected, the active input area will be indicated by a thin black frame in layout window.
• Algorithm controls:
  • Buttons:
    • Scramble: Generates a random layout of the current graph.
    • Shake: Applies a single random jump with the probability and maximum jump length defined in the parameter controls to every node.
    • Start: Starts or restarts the learning/layout algorithm (without initializing).
    • Stop: Stops or interrupts the learning/layout algorithm.
    • Initialize: Resets all algorithm paramters to the values given in the parameter control window. Also reset the stimulus area to the maximal area possible in the current window and with the correct shape.
    • Save: "Saves" the current graph with layout and paramters. Since this is a Java 1.0.X applet, real saving is not possible. Instead a textual description which can be stored in a savefile and loaded again is written to the Java console.
    • Load: Loads a new graph and its layout parameters from an ascii-file with the structure that is generated by "save". Since this is a Java 1.0.X applet all load-files must reside on the same server (and in the same directory) as the applet code. So you may have to download the applet if you want to use this funtion. The filename of the load-file must be entered in the "input" field before pressing the button. Hint: Entering an invalid filename or leaving the filename blank and then pressing "load" can be used clear the editor window.
    • Controls: Opens the parameter control window or brings it to the foreground.
  • Checkboxes:
    • Clip: If this box is checked, only nodes in the currently active stimulus area (see above) will be updated. Nodes outside the stimulus area remain unchanged even if they belong to the relevant topological neighborhood.
    • Taboo: (Obsolete) For testing taboo areas. When this box is checked each node updated will be checked for violation of another node's taboo area as defined in the parameter controls. If it penetrates a foreign taboo area it will bounce back along its update trajectory just to the border of this taboo area. This mode was used to test how a simple (and relatively inexpensive) taboo check would behave which is not required to be complete. This functionality does not seem to improve the layout.
    • Random Jumps: (Obsolete) While this box is checked a single random jump with the probability and maximum jump length defined in the parameter controls is applied to every node in each epoch. This was used to test whether the system could be driven to a different attactor by using random jumps. (Note: There is no cooling used for the random jump probability, but the functionality can be switched on and of interactively while the algorithm is running.) This functionality does not seem to improve the layout and is obsolete.
    • Trace:If checked trace information about current radius, current adaption factor, and epoch number is generated in the Java console window while the algorithm is running.
  • Input Field: Used for various interactive text inputs in conjunction with functions such as generating a new graph node (enter the label here) or loading a file (enter the filename here).

The other window is the parameter control window.

Here is a snapshot of this window which consists of one pop-up selector and a number of sliders described below.

• Distribution: Slects the shape of the active area for input stimuli. The area proper is defined interactively by using the area mode of the graph editor in the main window.
• Max. Epochs: The number of epochs (stimuli) after which the learning is stopped.
• Max. Radius: The initial radius of the topological neighborhood to be updated.
• Min. Radius: The minimum radius of the topological neighborhood to be updated at which the narrowing is stopped.
• Narrowing Interval: The number of epochs (stimuli) for which the radius is held constant.
• Max. Adaption: The initial adaption factor alpha.
• Min. Adaption: The minimum adaption factor alpha at which the cooling is stopped.
• Cooling Factor: The cooling factor c.
• Taboo Size: (Obsolete) The size of the taboo area around each node in pixel (see above)
• Max. Jump Radius: (Obsolete) The maximum length of a node's random jump (see above).
• Temperature: (Obsolete) The probability of a node to perform a random jump (see above).
• Delay: Applies a delay of X msecs. between sucessive epochs so that the animation can be watched more easily.
  Caution: On some older JVMs setting the delay to 0 msec. might cause threading problems.