Visualizations

Introduction

SOM Option Pack contains various visualization methods for analysing different aspects of SOM. Usually these are used together so that the same SOM is visualized using several visualizers. It is also possible to show the SOM data (or one or some of the feature vectors) in traditionel charts. Current visualizer implementations include:

  1. Colorized Sammon's Mapping - for overview and cluster detection
  2. Unified Matrix - for overview and for determining cluster boundaries
  3. Component Planes - for analysing the distribution of one variable over the map
  4. Chart Diagram of One Feature Vector - for comparing cells / map areas

Colorized Sammon's Mapping

Sammon's Mapping shows the relative distances between codebook vectors. The colorized version is made by spanning the Sammon's mapping of the SOM into a carefully selected color plane. The color plane is a selected plane of CIELab color space containing colors which can be easily separated by human eye. In the coloring, those cells which are close to each other will get the same or nearly similar color. If there is a big difference between two adjanced cells, the color changes radically between those cells. Thus if there are well separated clusters in the data, they can be differentiated easily by their colors. However, if there are no clear cluster boundaries in the data, the color fades smoothly from one to another.

Unified Matrix

Unified Matrix visualization displays the distances between map cells. The distances are shown by additional nodes in the visualized map. The coloring of those nodes describes the distance between the cells they connect. Usually the darker color means bigger distance. The cells itself are colored with the color correspondig the average distance from that cell. If there are well separated clusters in the map, they are "float" areas with clear boundaries.

Component Planes

The distribution of values of one variable over the map can be shown with component plane visualization. There the cell colors are selected according to the value of the variable. The color scale can be selected as in other charts.

Feature Vector Visualization

Each of the cells have got a feature vector during training. That can be displayed using the traditionel diagram charts. To select the proper cell, there is a <somCellFilter> data source.

Visualization Attributes

As Self-Organized Maps are visualized using Davisor Chart, nearly all of the normal chart rendering attributes can be used. The picture below shows how the most common rendering options are controlled.

SOM uses some of the attributes for delivering SOM specific information to the chart drawing module. In those cases it just ignores the values set by user. These overridden attributes include: