More information about data set visualization demo

The controls

Here are some more detailed instructions for the second demo:

Input data classes

The four coloured "pie" shapes on the left define the input classes. There are four classes represented by the colours red, yellow, green and blue. In each pie the sector represents the distribution of vectors that the class produces. All pies produce random vectors that are evenly distributed within the sector shown in the pie. The pies can be edited with the mouse in two ways: dragging the mouse near the center of the pie rotates the whole sector. Using the mouse near the edge of the circle changes the 'width' of the sector.

Proportions of classes

The vertical bar next to the pies sets the relative proportions of the four classes in the input data. Initially the input vectors are distributed equally among all classes. You can edit the proportions with the mouse: press down the mouse button to grab hold of the nearest color border, then move the border where you want with the mouse. The red area corresponds to the red pie on the left, etc.

The map

The area on the right is the self-organizing map. The map consists of 20 map units organized as a 5x4 unit grid (rectangular topology). The map is taught at the rate of 20 iterations per second and the map is classified and redisplayed once a second. In the classification, the background color of the units are set to the color of the input class they are closest to.

Each map unit contains a two-dimensional model vector whose components are interpreted as two coordinates, x and y. Each map unit is represented by an arrow that points from the coordinates [0,0] to coordinates [x,y] stored in the model vector. The range of the coordinates is from -1 to 1. The background color of the map unit tells which input class it represents the best (see the pies on the left). The model vectors in the map units (the black arrows) can be edited with the mouse.

Alpha/radius controls

Below the pies and the map there is the control panel that controls the alpha (learning rate) and neighborhood radius parameters of the algorithm. The upper row controls alpha and the lower controls radius. They both work in the same way: On the left the text field shows the initial value of the parameter which can be changed. Simply enter the new value and press return. The next display shows what the value of the parameter is relative to the starting and the ending values. The number displayed shows the actual value of the parameter. The red vertical line shows the position (use mouse to change its position). Alpha runs from the initial value to zero and radius runs from the initial value to 1. By default, when the teaching is in progress, the alpha and radius parameters decrease over time. You can stop them to constant value with the STOP button. To set it moving again, press the button again (it should read START). RESET resets the value of the parameter to the initial value (and moves the red indicator to the initial, leftmost position).

Main controls

Finally, the control panel in the bottom is used to start and stop teaching, randomize the map and to reset the values. Press START to start teaching the map, press it again to stop. Pressing the RANDOMIZE MAP button initializes the map to random values. RESET resets the alpha and radius parameters described above.
* Demo Index * Neural Networks Research Centre *

Last modified: Mon Apr 28 14:54:46 1997